As of: 2026-10-25 23:49 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/topics/product/

# Products: everything we have

81 events (67 after June 2026), 4 models, 96 videos, 48 posts. Latest event 22 October 2026.

Importance: * minor to ***** historic. Confidence says how well the event is sourced. Each event has its own page at https://postcutoff.com/e/<id>/.

## Key moments

- 2021-06-29: GitHub Copilot and OpenAI Codex bring LLMs to programming (****) https://postcutoff.com/e/2021-06-29-github-copilot-codex/
- 2022-11-30: OpenAI launches ChatGPT (*****) https://postcutoff.com/e/2022-11-30-chatgpt/
- 2026-01-12: Anthropic launches Claude Cowork — "Claude Code for the rest of your work" (****) https://postcutoff.com/e/2026-01-12-claude-cowork/
- 2026-06-08: WWDC 2026: Apple unveils Siri AI and new Apple Foundation Models built with Google's Gemini (****) https://postcutoff.com/e/2026-06-08-wwdc-2026-siri-ai-gemini/
- 2026-09-14: Apple ships iOS 27 with Gemini-assisted "Siri AI" after unveiling the 2nm A20 Pro iPhone 18 Pro (****) https://postcutoff.com/e/2026-09-14-ios-27-siri-ai-release/
- 2026-09-23: Meta Connect 2026: VR Glasses, Ray-Ban Meta Gen 3, camera-free audio glasses and Muse everywhere (****) https://postcutoff.com/e/2026-09-23-meta-connect-2026/
- 2026-09-25: Microsoft unveils the 'new Copilot' with Home, Code and Autopilot agents, offering Astra and Fable models (****) https://postcutoff.com/e/2026-09-25-microsoft-new-copilot-home-code-autopilot/
- 2026-09-28: Meta launches an Enterprise Platform division led by ex-MongoDB CEO CJ Desai, and Muse for Small Business (****) https://postcutoff.com/e/2026-09-28-meta-enterprise-platform-muse-business/
- 2026-09-29: OpenAI DevDay 2026: dots agents, GPT-6.1 Sol, Ultrafast, a $500 Pro plan and 20+ launches (****) https://postcutoff.com/e/2026-09-29-openai-devday-2026/
- 2026-10-07: OpenAI brings GPT-6 to all ChatGPT users with 'Intelligent UI' answers built from interactive components (****) https://postcutoff.com/e/2026-10-07-gpt-6-chatgpt-intelligent-ui/

## Timeline, oldest first

### 2021-06-29: GitHub Copilot and OpenAI Codex bring LLMs to programming

****; GitHub, OpenAI, Microsoft; Products; confidence high; https://postcutoff.com/e/2021-06-29-github-copilot-codex/

GitHub launched Copilot as a technical preview, an AI pair programmer powered by OpenAI Codex, a GPT model fine-tuned on public code; the Codex paper introduced the HumanEval benchmark.

- Copilot technical preview announced 29 June 2021
- Codex paper 'Evaluating Large Language Models Trained on Code', arXiv 2107.03374 (July 2021)
- Introduced HumanEval (164 hand-written Python problems)
- Copilot became generally available in June 2022

#### What happened
Copilot offered inline code completions in editors, generating whole functions from comments and context.

#### Why it matters
The first mass-market generative AI product for professionals; coding became the flagship LLM use case, leading to coding agents like Claude Code.

Sources: [Evaluating Large Language Models Trained on Code (arXiv)](https://arxiv.org/abs/2107.03374); [Introducing GitHub Copilot: your AI pair programmer (GitHub Blog)](https://github.blog/news-insights/product-news/introducing-github-copilot-ai-pair-programmer/); [Wikipedia: GitHub Copilot](https://en.wikipedia.org/wiki/GitHub_Copilot)

### 2022-11-30: OpenAI launches ChatGPT

*****; OpenAI; Products; confidence high; https://postcutoff.com/e/2022-11-30-chatgpt/

OpenAI released ChatGPT, a conversational interface to a GPT-3.5 model fine-tuned with RLHF, as a free research preview; it became the fastest-growing consumer app to that point and triggered the generative AI boom.

- Launched 30 November 2022 as a free research preview
- Based on a model in the GPT-3.5 series, trained with RLHF
- Passed 1 million users within about five days
- Estimated at ~100M monthly users by January 2023 (UBS/Similarweb estimate)
- ChatGPT Plus ($20/month) launched February 2023

#### What happened
ChatGPT let anyone chat with a capable LLM for free; its viral success forced Google, Meta, Microsoft and others into an AI race.

#### Why it matters
The moment AI became a mass-market technology — the start of the current era of AI investment, adoption and policy attention.

Sources: [Introducing ChatGPT (OpenAI)](https://openai.com/index/chatgpt/); [Wikipedia: ChatGPT](https://en.wikipedia.org/wiki/ChatGPT)

### 2025-05-19: Microsoft unveils Discovery, an agentic R&D platform, and says it found a non-PFAS datacenter coolant in ~200 hours

**; Microsoft; Products; confidence medium; https://postcutoff.com/e/2025-05-19-microsoft-discovery/

At Build 2025 (19 May 2025) Microsoft announced Microsoft Discovery, an enterprise agentic AI platform for scientific R&D on Azure. As a showcase, Microsoft said its researchers used the platform's models and HPC simulation to find a novel non-PFAS immersion coolant prototype in about 200 hours and synthesized it in under four months. No paper has been published on the coolant. Discovery reached general availability at Build 2026 (2 June 2026).

- Announced at Microsoft Build, 19 May 2025, as an enterprise agentic platform built on Azure with a graph-based knowledge engine
- Coolant case study: ~367,000 candidates screened; a non-PFAS immersion-coolant prototype found in ~200 hours of AI and HPC work, synthesized in under 4 months; Microsoft says measured properties matched predictions (company claim, no peer-reviewed paper)
- General availability announced 2 June 2026 (Aseem Datar), plus a preview desktop Discovery app on GitHub (github.com/microsoft/discovery)
- Named users: Yale Engineering, Georgia Tech, PNNL, Ginkgo Bioworks, GSK, BHP, Syensqo, Wiley; no pricing disclosed

#### What happened
Microsoft Discovery lets R&D teams run specialised AI agents over their own knowledge, simulation tools and experimental data. At launch Microsoft showed an internal case study. Its models and HPC simulations screened hundreds of thousands of candidate molecules for a PFAS-free immersion coolant for datacenters, the lab synthesized a prototype, and a PC was run submerged in it. A year later, at Build 2026, the platform became generally available and got a local desktop app in preview.

#### Why it matters
Microsoft Discovery is Microsoft's answer to Google's and Anthropic's AI-for-science products, and a continuation of its earlier battery-electrolyte screening with PNNL. The coolant claim has not been independently verified or published in a peer-reviewed venue.

Sources: [Azure blog: Transforming R&D with agentic AI, introducing Microsoft Discovery](https://azure.microsoft.com/en-us/blog/transforming-rd-with-agentic-ai-introducing-microsoft-discovery/); [Azure blog: Microsoft Discovery general availability and app preview (2 June 2026)](https://azure.microsoft.com/en-us/blog/announcing-microsoft-discovery-general-availability-and-microsoft-discovery-app-preview/); [VentureBeat: Microsoft AI discovered a new chemical in 200 hours](https://venturebeat.com/ai/microsoft-just-launched-an-ai-that-discovered-a-new-chemical-in-200-hours-instead-of-years); [PCWorld: Microsoft used AI to invent a safer coolant and dunked a PC in it](https://www.pcworld.com/article/2787517/microsoft-used-ai-to-invent-a-safer-coolant-and-dunked-a-pc-in-it.html); [Redmondmag: Build 2026, Microsoft Discovery hits GA](https://redmondmag.com/articles/2026/06/02/microsoft-discovery-hits-ga.aspx)

### 2025-10-27: xAI launches Grokipedia, an AI-written encyclopedia meant to rival Wikipedia

***; xAI; Products; confidence high; https://postcutoff.com/e/2025-10-27-grokipedia-launch/

On 2025-10-27 xAI launched Grokipedia v0.1, an online encyclopedia of about 885,000 articles generated by Grok and not editable by the public. Elon Musk pitched it as a less biased alternative to Wikipedia. Critics found many articles copied from Wikipedia and others pushing misinformation and far-right framing. Wikipedia editors deprecated it as a source by February 2026.

- v0.1 launched 2025-10-27 with ~885,000 Grok-generated articles; v0.2 on 2025-11-21; over 5.6 million articles by early 2026 (Wikipedia)
- Users cannot edit directly; they can suggest corrections through a form, and xAI controls the content
- Traffic peaked at 460,000+ US daily visits on 2025-10-28, then fell to about 35,000/day by mid-November
- Many articles were adapted from Wikipedia, some near-verbatim with a CC BY-SA notice
- Analyses found HIV/AIDS denialism, vaccine–autism claims, climate denial and white-nationalist framing (e.g. a Guardian investigation)
- From January 2026 some other chatbots (GPT-5.2, Google AI Overviews, Copilot) were seen citing Grokipedia; Wikipedia deprecated it as unreliable by February 2026
- Edit freeze confirmed: Lawfare (Aug 2026) found Grokipedia had not updated any article since April 24, 2026, with 13,002 suggested edits stuck 'in review' and no announcement, while the site still drew ~6.7M visits in June
- Late September 2026: updates resumed (user edits processed again, fresh Grok fact-checks, a design refresh); X/SpaceXAI design head Benji Taylor said the team had not forgotten Grokipedia; SpaceXAI did not comment to The Verge

#### What happened
xAI put Grok to work writing a whole encyclopedia and released it as Grokipedia. It started with roughly 885,000 articles and
grew into the millions within months.

#### Why it matters
It was the first large attempt to replace a human-edited reference work with a model-written one. It shows a feedback risk:
AI-written reference pages get cited back by other AI systems. The 2026 facts above (article counts, citation by other
chatbots, freeze in April 2026) come from the Wikipedia article and were not checked against primary sources, so treat them
as medium confidence.

Sources: [Lawfare: Grokipedia stopped reviewing edits in April. It didn't tell anyone.](https://www.lawfaremedia.org/article/grokipedia-stopped-reviewing-edits-in-april.-it-didn-t-tell-anyone); [PPC Land: Grokipedia halts edits on April 24, leaving 13,002 suggestions stuck](https://ppc.land/grokipedia-halts-edits-on-april-24-leaving-13-002-suggestions-stuck/); [iTechPost: Grokipedia is updating again (Sept 30, 2026)](https://www.itechpost.com/articles/237453/20260930/elon-musks-grokipedia-updating-again-latest-ai-encyclopedia-update.htm); [Grokipedia](https://grokipedia.com/); [Wikipedia - Grokipedia](https://en.wikipedia.org/wiki/Grokipedia); [MLQ - xAI launches Grokipedia](https://mlq.ai/news/elon-musks-xai-launches-grokipedia-open-source-ai-encyclopedia-aiming-to-rival-wikipedia/)

### 2026-01-12: Anthropic launches Claude Cowork — "Claude Code for the rest of your work"

****; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-01-12-claude-cowork/

On January 12, 2026 Anthropic launched Claude Cowork as a research preview in the Claude Desktop macOS app. It is a general agent for non-developers: it works in user-granted local folders, plans, splits tasks into parallel subtasks, and delivers finished files such as spreadsheets, decks and documents. It reached Pro users on Jan 16 and general availability on April 9.

- Research preview Jan 12, 2026 for Max subscribers; Pro access from Jan 16
- Built on the Claude Code agent harness, with a visual interface in Claude Desktop
- General availability around April 9, 2026 with enterprise features
- Merged with regular chat into 'one Claude' on Sept 16, 2026

#### What happened
Anthropic says Cowork grew out of users repurposing Claude Code for non-coding tasks. It pulls from local files, cloud tools and the web, and it can run several tasks at once.

#### Why it matters
It took Claude Code's agentic pattern to mainstream knowledge work, and it became one of Anthropic's biggest product lines of 2026.

Sources: [Simon Willison: First impressions of Claude Cowork](https://simonwillison.net/2026/Jan/12/claude-cowork/); [Axios: Anthropic's Claude moves further into the cubicle](https://www.axios.com/2026/01/12/ai-anthropic-claude-jobs); [Introducing Cowork (Anthropic video)](https://www.youtube.com/watch?v=UAmKyyZ-b9E)

### 2026-03-25: Claude Code auto mode: model-based classifiers approve tool calls instead of the user

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-03-25-claude-code-auto-mode/

Anthropic described Claude Code's auto mode (Mar 25, 2026), which hands permission decisions to a two-stage transcript classifier as a middle ground between approving every action and --dangerously-skip-permissions. It reported a 0.4% false-positive rate and a 17% false-negative rate on real "overeager" actions.

- Motivation: users approved 93% of permission prompts (approval fatigue)
- Two-stage transcript classifier; 0.4% false-positive rate after both stages
- False negatives: 17% on real overeager actions, 5.7% on a synthetic data-exfiltration set
- Evaluated on ~10,000 tool calls from real internal traffic
- Escalates to the user after 3 consecutive or 20 total denials

#### What happened
In the post (John Hughes), Anthropic calls auto mode "a new mode for Claude Code that delegates approvals to model-based
classifiers—a middle ground between manual review and no guardrails." The published error rates show it misses about one in six
real overeager actions.

#### Why it matters
Auto mode made classifier-gated autonomy the default way many developers ran coding agents in 2026, and its published
false-negative rate is a baseline for later incidents where agents acted beyond their instructions.

Sources: [Anthropic Engineering: Beyond permission prompts, making Claude Code more secure and autonomous (sandboxing, Oct 20, 2025)](https://www.anthropic.com/engineering/claude-code-sandboxing); [Anthropic Engineering: How we built Claude Code auto mode](https://www.anthropic.com/engineering/claude-code-auto-mode)

### 2026-03-31: Claude Code source code leaks via a source-map file in the npm package

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-03-31-claude-code-source-leak/

On March 31, 2026 Anthropic accidentally published the full Claude Code source, more than 512,000 lines of TypeScript in about 1,900 files, inside npm package v2.1.88 through a 59.8 MB source-map file. The leak exposed unreleased feature flags, including an always-on background agent called KAIROS. Anthropic called it a packaging error caused by human error, not a security breach.

- Date: March 31, 2026; @anthropic-ai/claude-code v2.1.88 shipped cli.js.map (59.8 MB)
- 512,000+ lines of TypeScript across 1,906 files; 44 hidden feature flags reported
- Discovered by security researcher Chaofan Shou; post reportedly drew 16–21M views
- GitHub disabled more than 8,100 mirror repositories

#### What happened
The root cause was reportedly a missing `*.map` exclusion in `.npmignore`. The leak revealed upcoming features and model references. Anthropic pulled the package.

#### Why it matters
It was a rare look at the internals of the most widely used AI coding agent. It came days after the Mythos draft leak (Mar 26) and raised questions about Anthropic's operational security.

Sources: [InfoQ: Claude Code source leak](https://infoq.com/news/2026/04/claude-code-source-leak); [DEV Community: The great Claude Code leak of 2026](https://dev.to/varshithvhegde/the-great-claude-code-leak-of-2026-accident-incompetence-or-the-best-pr-stunt-in-ai-history-3igm); [Penligent: Claude Code source map leak — what was exposed](https://www.penligent.ai/hackinglabs/claude-code-source-map-leak-what-was-exposed-and-what-it-means/)

### 2026-04-23: Anthropic postmortem: three changes degraded Claude Code quality in March–April 2026; usage limits reset

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-04-23-claude-code-quality-postmortem/

After weeks of user complaints that Claude Code had got worse, Anthropic published a postmortem (Apr 23, 2026) naming three causes: a default reasoning-effort cut from high to medium, a caching bug that wiped thinking history every turn, and a verbosity instruction that hurt coding. All were fixed by Apr 20; subscribers' usage limits were reset.

- Issue 1 (Mar 4 – Apr 7): default reasoning effort lowered from high to medium for Sonnet 4.6 and Opus 4.6 to cut latency; reverted
- Issue 2 (Mar 26 – Apr 10, fixed in v2.1.101): thinking-cache bug cleared reasoning history on every turn, making Claude 'seem forgetful and repetitive'
- Issue 3 (Apr 16 – Apr 20, fixed in v2.1.116): system-prompt word-limit instruction hurt coding quality (Sonnet 4.6, Opus 4.6, Opus 4.7)
- API and inference layer unaffected: 'We never intentionally degrade our models'
- Usage limits reset for all subscribers on Apr 23; promised per-model evals, ablations, gradual rollouts and soak periods for prompt changes

#### What happened
Anthropic wrote: "The implementation had a bug. Instead of clearing thinking history once, it cleared it on every turn for the rest
of the session." Together with the effort change and the prompt change, this explained much of the reported degradation.

#### Why it matters
It confirmed that "the model got dumber" complaints can come from harness, prompt and default changes rather than the weights,
and set a precedent for public postmortems on agent product quality.

Sources: [Anthropic Engineering: An update on recent Claude Code quality reports](https://www.anthropic.com/engineering/april-23-postmortem)

### 2026-05-06: Code with Claude 2026: Managed Agents "dreaming", doubled Claude Code limits and SpaceX Colossus 1 compute deal

***; Anthropic; Products; confidence medium; https://postcutoff.com/e/2026-05-06-code-with-claude-2026/

Anthropic's second Code with Claude developer conference (San Francisco, May 6–7, 2026; London May 19; Tokyo June 10) brought new Managed Agents capabilities (dreaming, outcomes, multi-agent orchestration), doubled Claude Code five-hour rate limits, and, per third-party recaps, a compute deal to use all of SpaceX's Colossus 1 data center (220,000+ NVIDIA GPUs, 300+ MW).

- San Francisco May 6 (plus indie/founder day), London May 19, Tokyo June 10
- Managed Agents: 'dreaming' (agents rehearse on past data), outcomes, multi-agent orchestration
- Claude Code five-hour rate limits doubled across Pro, Max, Team, Enterprise; peak-hour throttle lifted
- Reported SpaceX Colossus 1 compute partnership: >220,000 NVIDIA GPUs, >300 MW (third-party recap, not verified from primary source)
- Later (X reply, ~July 10, 2026, per Outlook Business): asked whether SpaceXAI could hurt Anthropic by ending the Colossus 1 lease (~$1.25B/month through May 2029, 90 days' notice, per SpaceX's S-1 via Business Insider), Musk wrote 'I was clearly wrong about Anthropic. They are obviously currently the leader in AI' and 'I would never cut them off in a way that hurt them badly, even as a competitor'; the quote resurfaced in Stocktwits headlines on Oct 17, 2026

#### What happened
Other May launches around the conference, per a third-party timeline: a Skills marketplace (~600 skills, May 1), Claude Platform on AWS GA (May 11), the Claude Code `/goal` command (May 12), and the acquisition of Stainless (SDK tooling, May 18).

#### Why it matters
The conference marked Anthropic's shift toward hosted agents and showed how much compute it was lining up, including from Elon Musk's SpaceX/xAI infrastructure.

Sources: [Code with Claude (Anthropic event page)](https://www.anthropic.com/events/code-with-claude); [Apito: Code with Claude recap — Managed Agents, SpaceX compute, doubled limits](https://apito.ai/en/blog/news/code-with-claude-conference/); [Dotzlaw Consulting: Anthropic's 2026 Code with Claude](https://dotzlaw.com/insights/anthropic-2026-code-with-claude/); [Outlook Business: Elon Musk admits he was wrong about Anthropic, calls it the leader in AI (July 10, 2026)](https://www.outlookbusiness.com/deeptech/elon-musk-admits-he-was-wrong-about-anthropic-calls-it-the-leader-in-ai-industry)

### 2026-05-19: Google launches 'Gemini for Science' at I/O 2026: Co-Scientist, AlphaEvolve and ERA become products

***; Google, Google DeepMind, Google Research; Products; confidence high; https://postcutoff.com/e/2026-05-19-google-gemini-for-science/

At Google I/O on 19 May 2026, Google bundled its science-research systems into 'Gemini for Science'. It has three experimental Google Labs tools: Hypothesis Generation (built on Co-Scientist), Computational Discovery (built on AlphaEvolve and Empirical Research Assistance, ERA) and Literature Insights (built on NotebookLM). It also added a science skills bundle for Antigravity, and Co-Scientist and AlphaEvolve for enterprises in private preview on Google Cloud. The same day, Nature published the ERA and Co-Scientist papers.

- Hypothesis Generation: multi-agent 'idea tournament' with cited, checked claims (labs.google/science)
- Computational Discovery: tests thousands of code variants in parallel (e.g. solar forecasting, epidemiology); gradual access through a trusted-tester program
- Literature Insights: turns papers into tables with custom searchable attributes, reports and audio/video summaries
- Science skills bundle for Google Antigravity: 30+ life-science databases incl. UniProt, AlphaFold DB, AlphaGenome API and InterPro
- Co-Scientist and AlphaEvolve in private preview for enterprise R&D on Google Cloud; no pricing disclosed
- ERA Nature paper ('An AI system to help scientists write expert-level empirical software'): LLM + tree search; 40 of 87 generated single-cell batch-integration methods beat every method on the OpenProblems v2.0.0 leaderboard (preprint arXiv 2509.06503, Sept 2025)
- ERA also reached or neared the top of CDC flu/COVID-19/RSV forecasting leaderboards and beat California's Bulletin 120 spring-runoff outlook, per Google Research
- Blog authors: Pushmeet Kohli (Google DeepMind / Google Cloud) and Yossi Matias (Google Research)

#### What happened
Google turned three research systems into products for scientists. Co-Scientist generates hypotheses. AlphaEvolve and ERA search over code to write better scientific software. NotebookLM handles the literature. They ship as Google Labs experiments and a science skills bundle for the Antigravity agent platform, and enterprise R&D teams get private previews on Google Cloud. Nature published the ERA and Co-Scientist papers the same day, alongside FutureHouse's Robin paper.

#### Why it matters
Google's "AI scientist" systems moved from research demos to products. This is the Gemini-centred strategy that later replaced dedicated single-problem teams such as AlphaFold's. The ERA leaderboard results are self-reported by Google, though they are now peer-reviewed.

Sources: [Google blog: Gemini for Science (I/O 2026)](https://blog.google/innovation-and-ai/technology/research/gemini-for-science-io-2026/); [Google Research: ERA, from Nature publication to computational discovery](https://research.google/blog/empirical-research-assistance-era-from-nature-publication-to-catalyzing-computational-discovery/); [Google Research at I/O 2026](https://research.google/blog/a-new-era-of-innovation-google-research-at-io-2026/); [Nature: An AI system to help scientists write expert-level empirical software (ERA)](https://www.nature.com/articles/s41586-026-10658-6); [arXiv 2509.06503 (ERA preprint)](https://arxiv.org/abs/2509.06503); [Nature: Accelerating scientific discovery with Co-Scientist](https://www.nature.com/articles/s41586-026-10644-y); [Google DeepMind: Co-Scientist, a multi-agent AI partner](https://deepmind.google/blog/co-scientist-a-multi-agent-ai-partner-to-accelerate-research/); [AIwire: Google pushes forward with new AI for Science tools](https://www.hpcwire.com/aiwire/2026/05/26/google-pushes-forward-with-new-ai-for-science-tools/)

### 2026-05-20: ElevenLabs launches Speech Engine: bring-your-own-LLM voice layer for existing chat agents

**; ElevenLabs; Products; confidence medium; https://postcutoff.com/e/2026-05-20-elevenlabs-speech-engine/

On 2026-05-20 ElevenLabs introduced Speech Engine, an API and SDK that turns an existing text chat agent into a voice agent. ElevenLabs handles transcription, TTS, turn-taking and interruption, while the developer's own server and LLM keep the conversation logic.

- Announced on X 2026-05-20: 'turn their existing chat agent into a full voice agent with one prompt'
- Combines ElevenLabs speech, transcription and voice-orchestration models in one pipeline; works with any LLM (OpenAI, Anthropic, Gemini, ...)
- WebSocket-based; JavaScript and Python SDKs manage connection lifecycle, turn-taking and interruption cancellation
- 70+ languages; SOC 2, HIPAA, GDPR, EU data residency, zero-retention mode (AlternativeTo summary)
- Pricing page lists burst pricing of $0.16/min; the standard per-minute rate ($0.08) is from a lead, not confirmed in our fetch
- 2026-09-21 changelog: new cascade_timeout_seconds parameter (2-15 s, default 4)

#### What happened
ElevenLabs split its voice-agent stack. ElevenAgents is the full hosted platform, and Speech Engine is a thin voice layer for teams that already have a text agent and want to keep their own LLM and logic.

#### Why it matters
It is the cascaded ("STT -> your LLM -> TTS") answer to end-to-end speech-to-speech models from OpenAI, Google and xAI. Developers keep full control of the model and tools and get ElevenLabs voices and turn-taking.

Sources: [ElevenLabs on X: Introducing Speech Engine](https://x.com/ElevenLabs/status/2057155693623361667); [ElevenLabs docs: Speech Engine](https://elevenlabs.io/docs/overview/capabilities/speech-engine); [ElevenLabs: Turn your chat agent into a voice agent](https://elevenlabs.io/speech-engine); [ElevenLabs API pricing](https://elevenlabs.io/pricing/api); [AlternativeTo: ElevenLabs launches Speech Engine](https://alternativeto.net/news/2026/5/elevenlabs-launches-speech-engine-for-instant-voice-integration-in-chat-agents/)

### 2026-05-28: Sesame launches its voice-companion iOS app (Maya, Miles, Simone, Charlie) in public preview

***; Sesame; Products; confidence high; https://postcutoff.com/e/2026-05-28-sesame-ios-app/

Sesame, the Oculus co-founders' conversational-voice startup behind the viral Maya/Miles demo and the open CSM-1B model, released a free public-preview iOS app on 2026-05-28 in 39 countries. It has four voice agents (Maya, Miles, Simone, Charlie), each with its own personality and memory. An Android preview is planned and smart glasses are targeted for 2027.

- Four agents: Maya, Miles, Simone, Charlie; 39 countries; free at launch; possible waitlist
- Company raised a $250M Series B (Oct 2025, Sequoia and Spark)
- Open model: sesame/csm-1b (Apache-2.0, March 2025)

#### What happened
After a year of web demos and a closed beta, Sesame shipped its natural-sounding voice companions as a consumer iPhone app.

#### Why it matters
Sesame's early-2025 demo reset expectations for conversational voice realism. The app tests whether voice-first companions become a daily habit before Sesame's planned glasses hardware.

Sources: [TechCrunch: Sesame launches its iOS app](https://techcrunch.com/2026/05/28/sesame-the-conversational-ai-startup-from-oculus-founders-launches-its-ios-app/); [Sesame](https://www.sesame.com/); [Hugging Face: sesame/csm-1b](https://huggingface.co/sesame/csm-1b)

### 2026-06-08: WWDC 2026: Apple unveils Siri AI and new Apple Foundation Models built with Google's Gemini

****; Apple, Google; Products; confidence high; https://postcutoff.com/e/2026-06-08-wwdc-2026-siri-ai-gemini/

At WWDC on 2026-06-08 Apple announced "Siri AI", a rebuilt conversational assistant with a standalone app, and a new generation of Apple Foundation Models developed in collaboration with Google's Gemini models (reportedly ~$1B/year deal). Developers got free Private Cloud Compute access and third-party model calls via the Foundation Models framework.

- Keynote 2026-06-08; iOS 27 and the other '27' OS releases announced
- Siri rebranded 'Siri AI': conversational, holds context, standalone app with chat history, cross-app actions
- Apple: next-generation Apple Foundation Models developed in collaboration with Google and the Gemini family
- Reported cost of Gemini deal: about $1 billion per year (secondary)
- AppleInsider: new foundation models 'don't contain a drop of Gemini' - Gemini used in development/training, not the shipped weights
- Free Foundation Models on Private Cloud Compute for developers with fewer than 2M first-time App Store downloads (MindStudio)
- Framework adds image input and access to third-party models such as Claude and Gemini via the same Swift API (MindStudio)
- iOS 27 supports iPhone 11 and later (TechCrunch)

#### What happened
Apple used WWDC 2026 to reset its AI strategy after the delayed 2024-25 Siri upgrade. It introduced **Siri AI** - a
conversational assistant with its own app, visual intelligence and cross-app task execution - and a new generation
of **Apple Foundation Models** developed in collaboration with Google's Gemini. Craig Federighi stressed that "privacy in
AI is non-negotiable", with requests processed on-device or in Private Cloud Compute. Other features: AI reply
suggestions in Messages, context-aware Phone app, system-wide AI dictation, generative Photos tools (Reframe, Extend,
Cleanup) and natural-language Shortcuts creation.

#### Why it matters
Apple, the largest consumer device platform, effectively conceded it could not build a frontier-class assistant alone
and partnered with Google, while keeping inference on its own privacy infrastructure.

Exactly how Gemini is used (training/distillation vs runtime) was reported inconsistently; AppleInsider and later
coverage say shipped models are Apple's own, trained with Gemini's help.

Sources: [TechCrunch - WWDC 2026: everything announced on Siri AI, iOS 27, Apple Intelligence](https://techcrunch.com/2026/06/09/wwdc-2026-everything-announced-on-siri-ai-os-27-apple-intelligence-and-more/); [AppleInsider - Apple's new foundation models don't contain a drop of Gemini](https://appleinsider.com/articles/26/06/08/apples-new-foundation-models-dont-contain-a-drop-of-gemini-as-we-said-they-wouldnt); [MacRumors - Apple outlines major AI and developer tool updates at Platforms State of the Union](https://www.macrumors.com/2026/06/09/apple-outlines-major-ai-and-developer-tool-updates/); [MindStudio - Apple Intelligence at WWDC 2026](https://www.mindstudio.ai/blog/apple-intelligence-wwdc-2026-ai-builders-guide)

### 2026-06-30: Anthropic launches Claude Science, an AI workbench for researchers (beta)

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-06-30-claude-science/

On June 30, 2026 Anthropic launched Claude Science in beta. It is a desktop workbench (macOS and Linux) that wraps existing Claude models in a research environment with 60+ scientific database integrations and a lead agent that delegates to specialized sub- agents. It launched with up to 50 grants of $30,000 in compute credits.

- Beta launched June 30, 2026 for Pro, Max, Team and Enterprise
- Not a new model; runs existing Claude models (e.g. Opus 4.8 at launch)
- 60+ scientific database integrations, focused on genomics and drug discovery
- Up to 50 projects to receive $30,000 in compute credits each (applications through July 15)
- Aug 27, 2026 expansion: 10,000 seats on a new Claude team plan for scientists for one year (standard seats free, premium seats with 5x limits $15/month) for PIs at academic or nonprofit institutions; AI for Science program widened beyond biology, up to $50,000 in credits per project
- Biology and chemistry users limited to Opus-class models at the time; Fable models block professional biology and drug-development queries; a US-government-partnered access program for Mythos-class life-sciences use had enrolled its first participants (apparently the Life Sciences Verification Program, which opened in public beta Sept 17 after enrolling its first organizations)

#### What happened
A main assistant acts as a research project manager: it organizes projects, connects data sources and delegates sub-tasks to specialized assistants.

#### Why it matters
It was Anthropic's first dedicated vertical product for science, a precursor to its wet lab and the Model Hardware Standard.

Sources: [Anthropic: Expanding our support for scientists (Aug 27, 2026)](https://www.anthropic.com/news/expanding-support-for-scientists); [Anthropic: Introducing the Life Sciences Verification Program (Sept 17, 2026)](https://www.anthropic.com/news/life-sciences-verification-program); [TechCrunch: Claude Science bets on workflow, not a new model](https://techcrunch.com/2026/06/30/anthropics-claude-science-bets-on-workflow-not-a-new-model-to-win-over-scientists/); [HPCwire/AIwire: Claude Science AI workbench](https://www.hpcwire.com/aiwire/2026/06/30/anthropic-launches-claude-science-ai-workbench-for-scientific-research/)

### 2026-07-01: xAI launches Grok Voice Agent Builder, a no-code platform for phone voice agents (beta)

**; xAI, SpaceX; Products; confidence medium; https://postcutoff.com/e/2026-07-01-xai-grok-voice-agent-builder/

On 2026-07-01 xAI (branded SpaceXAI) released the Grok Voice Agent Builder in beta: a browser-based, no-code tool that turns a plain-language description of a phone call into a live voice agent running on its single Grok Voice speech-to-speech model, with telephony, knowledge retrieval, tools/MCP, guardrails and call review bundled.

- Launch 2026-07-01, beta (x.ai news post); 'Create a personalized voice agent in under 2 minutes without a single line of code'
- Runs on one Grok Voice speech-to-speech model rather than a stitched STT -> LLM -> TTS pipeline
- Price: the x.ai post (read 2026-09-29) lists $0.08 per minute of audio (API rate) plus $0.01/min telephony on a provisioned number, no platform fee; Slator (2026-07-07) reported 'from $0.05 per minute'. The discrepancy is unresolved
- 25+ languages; voice cloning; integrations incl. Google/Outlook Calendar, email, web and X search, Linear, Notion, Google Drive, OneDrive; human transfer; SIP or phone-number deployment
- xAI-reported tau-voice Bench: Grok Voice Think Fast 1.0 67.3% vs Gemini 3.1 Flash Live 43.8% and GPT Realtime 1.5 35.3%
- Competes with ElevenLabs (ElevenAgents), Retell AI, Vapi, Synthflow and PolyAI (Slator)

#### What happened
xAI added a no-code layer on top of its Voice Agent API. Operators describe a call flow in plain language, attach documents and tools, test in the browser and deploy to a phone number or SIP trunk. Four weeks later (2026-07-29) the underlying model was upgraded to Grok Voice Think Fast 2.0.

#### Why it matters
Frontier labs moved into the voice-agent platform market that had belonged to ElevenLabs, Vapi and Retell. xAI's pitch was a single end-to-end speech model with telephony bundled, instead of a cascade built from several vendors. The per-minute price is unclear (see key facts), so confidence is medium.

Sources: [SpaceXAI: Introducing the Voice Agent Builder](https://x.ai/news/grok-voice-agent-builder); [SpaceXAI: Voice Agent Builder product page](https://x.ai/voice); [Slator: xAI Releases No-Code Voice Agent Builder](https://slator.com/xai-releases-no-code-voice-agent-builder/)

### 2026-07-23: Claude voice mode moves beyond Haiku to Opus and Sonnet, gains connectors and more languages

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-07-23-claude-voice-mode-opus-sonnet/

On 2026-07-23 Anthropic let Claude's voice mode run on Opus, Sonnet or Haiku (previously Haiku only), call connected tools mid-conversation (Gmail, Calendar, Slack, Canva, Notion) and speak more languages, in public beta on mobile, desktop and web. Anthropic still has no speech model or speech API of its own: voice mode remains a speech-to-text / text-to-speech wrapper whose provider is undisclosed.

- Voice mode uses the fastest version of the last model used in chat; model can be switched mid-conversation
- Free users: Haiku with one connected app; paid users: all three model families and multiple connectors
- Languages at launch included English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese (BR), Spanish
- Fable models are excluded from voice mode (Claude Help Center)
- No Anthropic TTS/STT or realtime audio API exists as of 2026-09-29; TTS/STT vendor not disclosed (TechCrunch)

#### What happened
Two weeks after OpenAI's full-duplex GPT-Live, Anthropic upgraded Claude's voice mode by letting its frontier models, not just
Haiku, answer spoken questions and act through connectors. The underlying cascaded voice pipeline was not replaced.

#### Why it matters
It shows the two strategies in voice: OpenAI and Google build native audio models, while Anthropic reuses its text models
with off-the-shelf speech components and competes on reasoning and tool use rather than conversational feel.

Sources: [Claude blog - Think through hard problems in voice mode](https://claude.com/blog/think-through-hard-problems-in-voice-mode); [Claude on X - voice conversations now use Opus and Sonnet](https://x.com/claudeai/status/2080376096873177300); [Claude Help Center - Use voice mode](https://support.claude.com/en/articles/11101966-use-voice-mode); [TechCrunch - Anthropic updates Claude voice mode with more capable models](https://techcrunch.com/2026/07/23/anthropic-updates-claude-voice-mode-with-more-capable-models/)

### 2026-07-25: Microsoft makes its own Azure Realtime speech-to-speech model generally available in the Voice Live API

**; Microsoft; Products; confidence high; https://postcutoff.com/e/2026-07-25-azure-realtime-voice-live/

On 2026-07-25 Microsoft made its in-house "Azure Realtime" speech-to-speech model (API id azure-realtime) generally available in the Azure Voice Live API. Microsoft says it is about 100 ms faster than GPT Realtime 1.5 and ships 34 locale-native voices in 11 languages. Voice Live itself is a managed speech-to-speech service, GA since November 2025, that wraps ASR (including MAI-Transcribe), an LLM (GPT-Realtime, GPT-5.x, Phi) and Azure TTS/avatars behind one Realtime-API-compatible WebSocket.

- Azure Realtime GA 2026-07-25: 34 locale-native voices across 11 languages; 'about 100 ms lower latency than GPT Realtime 1.5'; most voices 'on par with or better than competing offerings' (Microsoft)
- Voice Live API version 2026-07-15 GA (default for SDKs): 12 azure-realtime native voices, parallel tool calls, streaming text input, hosted-agent passthrough
- Voice Live service: GA November 2025; events mostly match the Azure OpenAI Realtime API; noise suppression, echo cancellation, semantic end-of-turn detection, avatars, function calling, MCP servers (GA April 2026)
- Model menu (Sept 2026): gpt-realtime-2.1 (+mini, datazone), gpt-realtime-1.5, gpt-5.6-terra/luna, gpt-5.x, gpt-4.1/4o, phi4-mm-realtime, azure-realtime; tiers Pro/Standard/Lite by model
- MAI-Transcribe is a preview speech-recognition option in Voice Live (since April 2026); MAI-Transcribe-2 and MAI-Voice-2 plug in as input/output

#### What happened
Microsoft had previewed an in-house speech-to-speech model ("Azure Realtime") around Build 2026 alongside the Voice Live
API. It reached GA in July 2026 as an alternative to OpenAI's gpt-realtime models inside Microsoft's managed voice-agent
service.

#### Why it matters
Microsoft now offers a first-party realtime voice model next to OpenAI's inside its own voice-agent platform. Together
with MAI-Transcribe and MAI-Voice, this is another sign that Microsoft is building a speech stack less dependent on
OpenAI. Per-minute pricing and independent benchmarks for azure-realtime were not found.

Sources: [Microsoft Learn - Voice Live release notes](https://github.com/MicrosoftDocs/azure-ai-docs/blob/main/articles/ai-services/speech-service/includes/release-notes/release-notes-voice-live.md); [Microsoft Learn - Voice Live API overview (models, pricing tiers)](https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live); [Microsoft Tech Community - Azure Speech at Build 2026: powering voice agents](https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/azure-speech-at-build-2026-powering-voice-agents-with-real-time-and-life-like-ex/4524638); [Microsoft Learn - MAI-Transcribe in Speech service](https://learn.microsoft.com/en-us/azure/ai-services/speech-service/mai-transcribe)

### 2026-08-06: Suno adds audio watermarking, fingerprinting and download limits amid lawsuits

**; Suno, Musixmatch; Products; confidence high; https://postcutoff.com/e/2026-08-06-suno-watermarking-download-limits/

Suno announced durable inaudible audio watermarks, fingerprinting (via Musixmatch's Sentinel copyright detection) and labels so its songs are identifiable on other platforms, banned deceptive "real" audio and unauthorized voice/likeness use, and then (2026-09-03) capped monthly downloads to curb mass uploads to streaming services and royalty fraud.

- Announced 2026-08-06 by CEO Mikey Shulman: tools 'designed to be durable and resistant to tampering, without affecting the listening experience'
- Partnership with Musixmatch for its Sentinel copyright-detection system
- Guidelines ban 'deceptive audio presented as real' and 'using a real person's voice or likeness without permission'
- Download limits from 2026-09-03 (ToS update): 20 songs/month on Pro, 60 on Premier; unlimited multitrack export from Suno Studio for Premier; free tier 7 lifetime downloads (per MBW)
- Suno also disclosed a November 2025 data breach affecting 55 million users (per TechCrunch)

#### What happened
A week after losing to GEMA in Munich, Suno rolled out provenance tools aimed mainly at streaming fraud (bulk-uploaded AI tracks boosted by bot plays) and impersonation, followed by per-plan download caps.

#### Why it matters
The largest AI music generator adopted watermarking and distribution limits voluntarily, ahead of EU AI Act transparency duties, as part of its pivot toward a licensed, label-friendly model.

Sources: [TechCrunch: Amid legal battles, Suno says it will start watermarking songs](https://techcrunch.com/2026/08/06/amid-legal-battles-suno-says-it-will-start-watermarking-songs/); [Engadget: Suno is adding audio watermarks](https://www.engadget.com/2231870/suno-adding-audio-watermarks-ai-generated-songs-identifiable/); [Suno: Terms of Service update (download limits)](https://suno.com/blog/suno-updates-tos); [MBW: Suno launches Studio 2.0 (download-limit table)](https://www.musicbusinessworldwide.com/suno-launches-studio-2-0-with-midi-support/); [MBW: Suno becomes first customer of Musixmatch's Sentinel](https://www.musicbusinessworldwide.com/suno-first-customer-of-musixmatchs-sentinel-which-screens-ai-prompts-and-outputs-for-copyrighted-material/)

### 2026-08-13: Suno Studio 2.0 adds MIDI, an AI chat bar that builds plugins, and stem separation to its browser DAW

**; Suno; Products; confidence high; https://postcutoff.com/e/2026-08-13-suno-studio-2/

Suno upgraded its browser-based generative audio workstation with MIDI recording/editing (MIDI clips can prompt new audio), a beta chat assistant that generates instruments and vocals and builds custom plugins and synth presets, a wavetable synth, better stem separation, effects, automation and unlimited 32-bit/48 kHz multitrack export, for Premier subscribers only.

- Launched 2026-08-13; Studio 1.0 had launched in beta on 2025-09-25
- MIDI clips usable as prompts for new generations; typing-keyboard play with arpeggiator and chord mode
- Beta chat bar can generate instruments/vocals and create new plugins and synth presets
- Unlimited 32-bit/48 kHz multitrack export (vs 20/60 monthly song downloads on Pro/Premier)
- Premier tier only ($24-30/month per MBW)

#### What happened
One day after Suno's BMG licensing deal, Suno shipped a major DAW update that blends conventional production tools (MIDI, synth, automation) with generative ones (chat-driven instrument and plugin generation).

#### Why it matters
It pushed Suno from a one-shot song generator toward a professional production tool, competing with DAWs rather than only with other generators, while steering heavy exporters to its top tier.

Sources: [Suno: Introducing Studio 2.0](https://suno.com/blog/studio-2); [Suno release notes: Studio 2.0 is here](https://suno.com/release-notes/studio-2); [Music Business Worldwide: Suno launches Studio 2.0 with MIDI support](https://www.musicbusinessworldwide.com/suno-launches-studio-2-0-with-midi-support/); [MusicRadar: Suno's Studio 2.0 adds an AI chatbot](https://www.musicradar.com/music-tech/sunos-studio-2-0-adds-an-ai-chatbot-that-can-control-your-project-transform-sounds-and-generate-custom-plugins)

### 2026-08-18: OpenAI launches 'ChatGPT for Teens' with age assurance, quiet hours and stricter content rules

***; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-08-18-openai-chatgpt-for-teens/

On Aug 18, 2026 OpenAI launched "ChatGPT for Teens", a separate experience for users aged 13–17. It has stricter limits on suicide, self-harm and romantic or sexual content, break nudges, Quiet Hours and Study Hours, and optional parental controls, all based on OpenAI's age-prediction system. It came amid lawsuits alleging ChatGPT conversations contributed to teen deaths, and age prediction reached the EU on Aug 25, 2026.

- Ages 13–17; ChatGPT barred from romantic language or terms of endearment with teens (CNN)
- Break nudge after 90 minutes of activity within a three-hour window; parents/teens can set Quiet Hours (ChatGPT unavailable) or Study Hours (Study Mode on by default) (CNN)
- Learning features: homework reminders, quizzes, learning visualizations; AI-literacy partnership (TechCrunch)
- Builds on age prediction, rolled out globally on consumer plans from Jan 20, 2026 (behavioural/account signals; adults misclassified can verify with a selfie)
- Age prediction began rolling out in the EU on Aug 25, 2026 (per OpenAI's age-prediction page, via search snippet)
- TechCrunch notes ChatGPT had about 900M weekly users before teen-specific protections
- What it is (Common Sense Media's description): not a separate app or model but a set of settings, prompts, classifiers and interface features switched on for accounts known or suspected to belong to 13–17-year-olds; it comes with an updated Under-18 section of OpenAI's Model Spec ('ChatGPT should not use romantic language, encourage emotional dependence, or imply that it has feelings or consciousness')
- Rollout: the eating-disorder parent notifications announced on Aug 18 went live on Sept 10, 2026 (Common Sense Media risk assessment)
- Oct 7, 2026: Common Sense Media's Youth AI Safety Institute rated ChatGPT for Teens an 'Unacceptable Risk' after 4,000+ test prompts (no parental alerts during hour-long self-harm chats on new linked accounts, Study mode easy to bypass, age prediction never triggered) and urged OpenAI to keep teens off ChatGPT; OpenAI disputed the methodology (see 2026-10-07-common-sense-media-chatgpt-for-teens-unacceptable-risk)

#### What happened
OpenAI had announced a teen version in September 2025 and age prediction in January 2026. The August 2026 launch combined them
into a distinct product experience with default-on restrictions and time-management tools.

On Oct 7, 2026 Common Sense Media published an independent test of the launch. It found that several promised features (parental
alerts, Study Hours, age prediction, less friend-like behaviour) did not work reliably, and rated the product an "Unacceptable
Risk"; OpenAI disputed the methodology (see 2026-10-07-common-sense-media-chatgpt-for-teens-unacceptable-risk).

#### Why it matters
It is the largest consumer chatbot's main response to the teen-safety lawsuits and regulatory pressure of 2025–26, and it uses
automated age estimation at very large scale, which raises privacy and accuracy questions.

Unverified: the OpenAI announcement post itself was not read (openai.com returned 403); details come from press coverage.

Sources: [CNN: OpenAI introduces 'ChatGPT for Teens' amid scrutiny over child safety](https://www.cnn.com/2026/08/18/tech/openai-chatgpt-for-teens); [TechCrunch: OpenAI launches a safer ChatGPT for teens](https://techcrunch.com/2026/08/18/openai-launches-a-safer-chatgpt-for-teens-years-after-teens-started-using-it/); [CNBC: OpenAI rolls out ChatGPT for Teens](https://www.cnbc.com/2026/08/18/openai-chatgpt-for-teens-safety.html); [Axios: OpenAI debuts ChatGPT for Teens](https://www.axios.com/2026/08/18/openai-chatgpt-for-teens); [OpenAI: Our approach to age prediction](https://openai.com/index/our-approach-to-age-prediction/); [Help Net Security: OpenAI adds age prediction to ChatGPT (Jan 21, 2026)](https://www.helpnetsecurity.com/2026/01/21/chatgpt-age-prediction-teen-safety/); [Common Sense Media Youth AI Safety Institute: ChatGPT for Teens risk assessment (Oct 7, 2026)](https://institute.commonsensemedia.org/risk-assessments/chatgpt-teens)

### 2026-09-10: Google launches a desktop Gemini app for Windows

**; Google; Products; confidence high; https://postcutoff.com/e/2026-09-10-gemini-app-windows/

On Sept 10, 2026 Google released a native Gemini app for Windows PCs, opened with the Alt + Space shortcut over any window. It can delegate multi-step tasks to Gemini Spark, work across Gmail and Drive files, and generate Nano Banana images or Gemini Omni videos.

- Download at gemini.google/desktop; Alt + Space opens Gemini over the current work
- Per Google's September recap: delegate multi-step tasks to Gemini Spark, synthesize files across Gmail and Drive, generate Nano Banana images or Gemini Omni videos
- Announced by Erin Pettigrew, Director of Product Management, Gemini App
- Oct 3 (unconfirmed, found by TestingCatalog): a hidden 'Additional sandbox options' setting in the Gemini Desktop app for Mac would let Gemini read, create, modify or delete files anywhere on the Mac and act through apps like Mail, Safari and Messages. Depending on the settings, it could act without asking first, but it would still confirm purchases, money transfers, account creation, accepting legal terms and changes to sensitive personal info (BleepingComputer). Google has not confirmed it

#### What happened
Google shipped a desktop Gemini app for Windows on Sept 10, 2026. A keyboard shortcut (Alt + Space) brings Gemini up over whatever the
user is doing. The app includes a workspace where users can hand multi-step tasks to the Gemini Spark agent and pull in Gmail and Drive
files, plus image and video generation.

#### Why it matters
It puts Google's assistant, and its Gemini Spark agent, directly on the Windows desktop alongside other AI assistants' desktop apps.

Sources: [BleepingComputer: Google Gemini could soon get full access to your Mac's files, apps and the web](https://www.bleepingcomputer.com/news/google/google-gemini-could-soon-get-full-access-to-your-macs-files-apps-and-the-web/); [Google: The Gemini app is now available for Windows](https://blog.google/innovation-and-ai/products/gemini-app/gemini-app-now-on-windows/); [Google: The latest AI news we announced in September 2026](https://blog.google/innovation-and-ai/technology/ai/google-ai-updates-september-2026/)

### 2026-09-11: OpenAI to retire custom GPTs on Dec 11, 2026 and migrate them to plugins (skills, reference files, apps and MCP tools); the GPT Store is phased out

***; OpenAI; Products; confidence medium; https://postcutoff.com/e/2026-09-11-openai-retires-custom-gpts-plugins/

On Sept 11, 2026 OpenAI told ChatGPT Enterprise admins that custom GPTs, introduced in Nov 2023, will be retired on Dec 11, 2026 (Feb 11, 2027 for Enterprise workspaces with an approved deferral) and replaced by plugins that work across ChatGPT and Codex. A "Migrate to plugin" button turns a GPT's instructions into a skill and its knowledge files into reference files; custom actions, the chosen model, sharing settings and past chats do not carry over. The GPT directory is phased out with them. Builders pushed back on the lost features.

- Notice to Enterprise admins and ChatGPT release notes: Sept 11, 2026; migration tool targeted for Sept 17; retirement Dec 11, 2026; approved Enterprise deferrals to Feb 11, 2027 (Mixed, Virtualization Review)
- End of new custom GPT creation for Enterprise: reported as Sept 25 (Mixed, Sept 18) and as Oct 26, 2026 (Virtualization Review, Sept 28); Free, Plus and Pro follow 'a similar timeline' with in-product notice
- Migration: My GPTs → 'Migrate to plugin'. Instructions become a skill, knowledge files become reference files, connected apps can be included; the new plugin starts private
- Not migrated: custom actions (must be rebuilt as connectors or MCP servers), selected model, conversation starters, capability settings, sharing settings and existing conversations; prior users get no automatic access
- OpenAI warns 'a migrated plugin may respond differently', so builders must retest prompts, file use, output formats and integrations
- GPT Store: the GPT directory is phased out as GPTs retire; migrated links should redirect to the replacement plugins for authorized users (Mixed)
- Context: at DevDay (Sept 29) OpenAI added plugin extensions with sidebar homes, panels and file viewers, a Plugin Creator and MCP Events support

#### What happened
OpenAI is folding custom GPTs, its 2023 way for users to package instructions, files and actions, into the newer plugin system shared with
Codex. Migration is one click for the simple parts, but API actions and sharing have to be rebuilt. The OpenAI help article was not readable
by our fetcher (403), so dates come from press summaries of it, which disagree on when Enterprise stops new GPT creation.

#### Why it matters
It ends the GPT Store experiment and makes skills, plugins and MCP the common way to extend ChatGPT and Codex, like the plugin and skill
systems at other labs. Every existing GPT must be migrated or will disappear by December.

Sources: [OpenAI Help Center: Custom GPT retirement and migration FAQ](https://help.openai.com/en/articles/20001519-custom-gpt-retirement-and-migration-faq); [Virtualization Review: OpenAI to retire custom GPTs, replace them with plugins (Sept 28)](https://virtualizationreview.com/articles/2026/09/28/openai-to-retire-custom-gpts-replace-them-with-plugins.aspx); [Visual Studio Magazine: Hands-on with OpenAI's mandatory custom GPT-to-plugin migration (Sept 28)](https://visualstudiomagazine.com/articles/2026/09/28/openai-replacing-custom-gpts-with-plugins-including-for-codex-in-vs-code.aspx); [Mixed: OpenAI is retiring custom GPTs, here's what carries over to plugins (Sept 18)](https://mixed-news.com/en/openai-retiring-custom-gpts-what-carries-over-to-plugins/); [OpenAI Developer Community: custom GPT → plugin migration raises access concerns](https://community.openai.com/t/openai-s-custom-gpt-plugin-migration-raises-access-concerns/1399984)

### 2026-09-14: Apple ships iOS 27 with Gemini-assisted "Siri AI" after unveiling the 2nm A20 Pro iPhone 18 Pro

****; Apple, Google; Products; confidence high; https://postcutoff.com/e/2026-09-14-ios-27-siri-ai-release/

Apple released iOS 27 worldwide on 2026-09-14, bringing the rebuilt Siri AI (opt-in beta, with daily usage limits and paid expanded access) to hundreds of millions of iPhones. Five days earlier, its 2026-09-09 event launched the iPhone 18 Pro with the A20 Pro - the first 2nm smartphone chip - and the foldable iPhone Duo.

- iOS 27 released 2026-09-14 as a free update
- Siri AI: opt-in beta, possible waitlist; daily usage limits with 'expanded access' for a fee (Apple fine print per MacRumors)
- Apple says it used Google's Gemini models to train the models behind Siri AI; inference runs on-device or in Private Cloud Compute, not via Gemini at runtime
- Apple claims Siri AI works with over 300,000 apps (CNBC live coverage)
- Apple event 'Surprise and Shine' on 2026-09-09
- A20 Pro: first 2nm smartphone chip; 6-core CPU, dual Neural Engines with 32 cores total, 50% more memory bandwidth (reported)
- iPhone 18 Pro: pre-orders Sept 12, launch Sept 18; iPhone Duo foldable from $1,999, launch Oct 23

#### What happened
On 2026-09-09 Apple introduced the iPhone 18 Pro/Pro Max with the **A20 Pro**, redesigned "desktop class" cores Apple
says make AI faster, built on TSMC's 2nm process, plus its first foldable, the **iPhone Duo**. On 2026-09-14 **iOS 27**
shipped, delivering the **Siri AI** experience announced at WWDC: a conversational assistant with a standalone app and
chat history, trained with help from Google's Gemini but running on-device or in Private Cloud Compute. It launched as
an opt-in beta with daily usage limits.

#### Why it matters
This is the moment Apple's long-delayed LLM Siri reached the mass market - the largest single rollout of a
frontier-derived assistant to existing devices - and the first time Apple has metered an AI feature with paid tiers.

A20 Pro core/Neural Engine specs come from secondary coverage.

Sources: [CNBC - Apple releases iOS 27, redesigned Siri AI](https://www.cnbc.com/2026/09/14/apple-releases-ios-27-redesigned-siri-ai.html); [CNBC - Apple event 2026 live updates](https://www.cnbc.com/2026/09/09/apple-event-today-live-updates.html); [MacRumors - Everything Apple announced at the September 2026 event](https://www.macrumors.com/2026/09/09/apple-september-2026-event-recap/); [Plain English - Apple ships Siri AI on iOS 27, built with Gemini, on 2nm A20 Pro](https://plainenglish.io/artificial-intelligence/apple-siri-ai-ios-27-gemini-a20-pro-september-2026); [Apple Event September 9 2026 (YouTube, Apple)](https://www.youtube.com/watch?v=39BalPDuTo0)

### 2026-09-15: Google ships Gemini 3.8 Live voice models and Gemini 3.8 Flash TTS with voice design and cloning

**; Google; Products; confidence high; https://postcutoff.com/e/2026-09-15-gemini-3-8-live-and-tts/

In September 2026 Google made its 3.8-generation audio models GA in the Gemini API: `gemini-3.8-live` and `gemini-3.8-live-extended-thinking` for real-time audio-to-audio agents (15 Sept), and `gemini-3.8-flash-tts` / `gemini-3.8-flash-lite-tts` plus a Voices endpoint with voice design and voice replication (22 Sept).

- 2026-09-15: gemini-3.8-live and gemini-3.8-live-extended-thinking GA (audio-to-audio, real-time)
- 2026-09-22: gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts GA
- New /v1beta/voices endpoint, voice design, voice replication and an Extended Voice Library
- Earlier: gemini-3.5-transcribe and gemini-3.5-transcribe-live GA on 2026-08-26; Lyria 3.5 music model GA on 2026-09-03
- Sept 24, 2026: Gemini 3.8 Live with Live Avatar generally available in Gemini Enterprise: near-real-time talking avatar (24 FPS video, 24 kHz audio, sub-second latency), lip-sync in 97 languages, asynchronous background tool calls, SynthID watermarking; US and EU endpoints; custom avatars need allowlisting
- Google (Sept 15): Gemini 3.8 Live Extended Thinking ranks #1 on Artificial Analysis' Speech-to-Speech Quality Index (82.6), scores 68.6% on a voice banking agent benchmark and 97.7% on Big Bench Audio; 3.8 Live handles near-real-time visual input, switches among 97 languages mid-conversation and runs tools in the background while talking
- Rolled out the same day in the Gemini app (Daily Brief, inbox), Google Workspace (Docs Live, Gmail Live, Keep Live) and Search Live; partners named: Salesforce, Genspark, Lumeris; output carries SynthID audio watermarks

#### What happened
Following Gemini 3.8 Flash, Google rolled the 3.8 generation into its real-time voice (Live) and text-to-speech models, adding APIs to design and replicate voices.

#### Why it matters
Completes a full voice stack (transcription, reasoning, real-time dialogue, speech synthesis, cloning) on one API; voice cloning also raises misuse concerns.

Sources: [Google: Gemini 3.8 Live with Live Avatar](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-with-live-avatar/); [Google Cloud: Gemini 3.8 Live with Live Avatar is now generally available](https://cloud.google.com/blog/products/ai-machine-learning/gemini-3-8-live-with-live-avatar-is-now-generally-available); [Gemini API release notes](https://ai.google.dev/gemini-api/docs/changelog); [Gemini API models overview](https://ai.google.dev/gemini-api/docs/models); [Google: Gemini 3.5 Transcribe](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5-transcribe/); [Google DeepMind: Gemini 3.8 text-to-speech says hello](https://deepmind.google/blog/say-hello-to-gemini-38-text-to-speech/); [Google DeepMind: Introducing Gemini 3.8 Live with Live Avatar](https://deepmind.google/blog/introducing-gemini-38-live-with-live-avatar/); [The Verge: Gemini Live Avatar gives the AI a face](https://www.theverge.com/tech/1000328/google-gemini-ai-live-avatar-face); [Simon Willison: Gemini 3.8 TTS Playground](https://simonwillison.net/2026/Sep/23/gemini-tts-playground/); [Google: Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking (Sept 15)](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/); [Google DeepMind: Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking](https://deepmind.google/blog/introducing-gemini-3-8-live-and-3-8-live-extended-thinking/); [Google blog: Gemini 3.8 text-to-speech says hello](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-text-to-speech/)

### 2026-09-16: Anthropic merges Cowork and chat into "one Claude" and launches Claude Docs, Slides and Design in beta

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-09-16-one-claude-docs-slides-design/

On September 16, 2026 Anthropic merged Claude Cowork and regular chat into a single Claude experience and launched Claude Docs and Claude Slides in beta, with Claude Design working inside conversations. Users can create, comment on and revise documents, decks and designs without leaving the chat. Projects were redesigned as a single conversation with parallel threads on Sept 17.

- Announced Sept 16, 2026
- Cowork, Claude Design and Artifacts modes unified under one chat
- Claude Docs exports to Word, PDF, Markdown and Google Docs; Claude Slides presents in Claude or exports PowerPoint/PDF
- Docs and Slides beta on paid plans, rolling out to Pro and Max first
- Claude Design first launched as a research preview April 17, 2026

#### What happened
Meaghan Choi, who leads design for Claude apps, explains in the official video why keeping bigger work in a separate place "stopped making sense". Chats, tasks, skills and memories stay where they were.

#### Why it matters
This is Anthropic's direct push into office productivity software against Microsoft 365 and Google Workspace.

Sources: [Computerworld: Anthropic launches Claude Docs and Slides](https://www.computerworld.com/article/4223177/anthropic-tries-to-make-claude-stickier-with-launch-of-docs-and-slides.html); [Meet Claude Slides, Claude Design and Claude Docs (video)](https://www.youtube.com/watch?v=To5nrYqvR44); [Claude Cowork and chat are now one Claude (video)](https://www.youtube.com/watch?v=qMUf-jwSpMo); [Projects are now a conversation with Claude (video)](https://www.youtube.com/watch?v=5qt_aGyAsKk)

### 2026-09-16: ElevenLabs launches Reception, an AI phone receptionist for small businesses built on ElevenAgents

**; ElevenLabs; Products; confidence high; https://postcutoff.com/e/2026-09-16-elevenlabs-reception/

On 2026-09-16 ElevenLabs launched Reception (reception.ai), a packaged AI receptionist for small businesses built on its ElevenAgents platform. It answers calls 24/7, answers questions about the business, books appointments and texts confirmations, and it is set up by adding the business's website.

- Announced 2026-09-16 (blog + X post x.com/ElevenLabs/status/2100262886916358361)
- Answers calls around the clock, answers questions, books appointments into a built-in or Google calendar, public booking page, takes messages
- Callers can speak 'in their own language' (the product page says 70+ languages)
- Plans from $22/month with a free trial (product page; pricing at reception.ai/pricing)
- ElevenLabs' first vertical, self-serve agent product aimed at non-developers

#### What happened
ElevenLabs packaged its agent platform as a turnkey product. A business owner points Reception at the company website, and it becomes a phone agent that handles inquiries and bookings.

#### Why it matters
Voice agents went from developer platforms to small-business subscriptions. A missed-call replacement at about $22/month competes directly with human answering services.

Sources: [ElevenLabs blog: Introducing Reception, an AI Receptionist by ElevenAgents](https://elevenlabs.io/blog/reception); [ElevenLabs on X: Introducing Reception](https://x.com/ElevenLabs/status/2100262886916358361); [Reception product page](https://elevenlabs.io/reception); [YouTube (ElevenLabs): Reception, powered by ElevenAgents](https://www.youtube.com/watch?v=3RojrjVVFSg); [Reception.ai docs](https://elevenlabs.io/docs/reception-ai/overview)

### 2026-09-17: Speechmatics launches Agent STT, powered by its Linden model, for voice agents

**; Speechmatics; Products; confidence high; https://postcutoff.com/e/2026-09-17-speechmatics-agent-stt-linden/

On 2026-09-17 Speechmatics launched Agent STT, a speech-to-text API built for production voice agents and powered by its new Linden 1 model. It returns speaker-attributed segments with turn events rather than a word stream, and Speechmatics reports a 1.05% semantic error rate and 369 ms median finalization on Pipecat's 23-model streaming STT benchmark. Launch price is $0.30/hour.

- Model: linden-1, served on a new /v2/agent endpoint; 55+ languages; segments finalized in under 350 ms
- Pipecat STT benchmark (vendor-cited): 1.05% pooled semantic error rate, 369 ms median finalization, on the speed/accuracy Pareto frontier of 23 streaming models
- Pricing: $0.30/hour at launch, $0.16/hour with volume discount
- Custom vocabulary up to 1,000 terms, live diarization and speaker ID; available via API, Pipecat and LiveKit
- Follows Melia 1 (2026-06-17), Speechmatics' code-switching multilingual batch model across 55+ languages

#### What happened
Speechmatics, the UK speech-recognition company, shipped a separate STT product for LLM voice agents. Its Linden 1 model is tuned for the errors that break calls:
a changed digit in an account number, a missed negation, a dropped one-word confirmation. Output comes as speaker-attributed segments with turn messages, ready to hand to an LLM.

#### Why it matters
Voice-agent STT is now a separate product category (Deepgram Flux, AssemblyAI Universal-3.x Pro Realtime, Cartesia Ink-2, Speechmatics Agent STT). Vendors compete on turn detection, semantic errors and finalization latency, not only average WER.
The benchmark numbers are Speechmatics' reading of Pipecat's public benchmark, not an independent audit.

Sources: [Speechmatics press release (GlobeNewswire): Agent STT](https://www.globenewswire.com/news-release/2026/09/17/3364138/0/en/speechmatics-launches-agent-stt-for-the-speech-errors-that-derail-voice-agents.html); [Speechmatics Agent STT product page](https://www.speechmatics.com/voice-agents); [Speechmatics docs: models (Linden 1, Melia 1)](https://docs.speechmatics.com/speech-to-text/models); [Speechmatics: Introducing Melia](https://www.speechmatics.com/company/articles-and-news/introducing-melia-multilingual-speech-to-text-model); [HackerNoon: Pipecat benchmarked 23 real-time STT models](https://hackernoon.com/pipecat-benchmarked-23-real-time-stt-models-for-voice-agents-there-isnt-one-winner)

### 2026-09-22: WHO AFRO, CEPI and DRC's INRB use Claude in the Bundibugyo Ebola outbreak response

***; Anthropic, WHO, CEPI; Products; confidence high; https://postcutoff.com/e/2026-09-22-claude-ebola-response-drc/

Anthropic's feature "The Situation Report" (Sept 22, 2026) describes how WHO's Africa regional office, CEPI and the DRC's INRB use Claude in the response to a Bundibugyo ebolavirus outbreak in eastern DRC (7,672 confirmed cases, 48.2% fatality). A Claude skill cut daily situation reports from a full day to under an hour. Teams also use Claude to run several epidemic models in parallel and to track vaccine R&D.

- Outbreak: Bundibugyo ebolavirus (BDBV) in eastern DRC, confirmed in May 2026; public health emergency of international concern declared on May 17; no approved BDBV vaccine
- Figures in the featured sitrep: 7,672 confirmed cases, 3,699 deaths (48.2% fatality), 7 provinces and 63 of 167 health zones; Ituri has 77.1% of cases
- WHO AFRO (Tendai Muza and colleagues) built a Claude skill that pulls case and lab numbers from each health zone's PowerPoint deck, checks them against the previous day and flags trend changes: sitreps went from all day to under an hour
- WHO AFRO's Paul Ouma: before, time allowed only one disease model; now they run several at once for forecasts, e.g. where to build treatment centres
- CEPI used Claude to build a dashboard tracking vaccine-development work (incl. Ervebo cross-reactivity against BDBV); CEPI says scientific judgements stay with experts
- Lab teams use Anthropic's scientific workbench for genome assembly and viral phylogenies (per the feature)
- Run by Anthropic's Beneficial Deployments and Applied AI teams in a partnership convened by CEPI

#### What happened
Anthropic published a long-form feature on how Claude fits into the data chain of an active Ebola response. Health workers' notebooks and
WhatsApp messages feed district PowerPoint decks, which feed the provincial situation report. Claude now does the consolidation and
consistency checks, and speeds up modelling and evidence review. Dr. Jean-Jacques Muyembe (INRB director general, co-discoverer of Ebola in
1976) is quoted: "you beat Ebola by knowing where it is today, not where it was last week."

#### Why it matters
It is a documented deployment of a frontier model inside a live, high-fatality outbreak response, with named WHO and CEPI users and concrete
time savings. The claims come from Anthropic's own feature, with no independent evaluation.

Sources: [Anthropic: The Situation Report (Ebola response feature)](https://www.anthropic.com/features/ebola-response)

### 2026-09-23: Meta Connect 2026: VR Glasses, Ray-Ban Meta Gen 3, camera-free audio glasses and Muse everywhere

****; Meta; Products; confidence high; https://postcutoff.com/e/2026-09-23-meta-connect-2026/

At Connect on 2026-09-23 Meta unveiled Meta VR Glasses (~100 g, $1,299.99, spring 2027), Ray-Ban Meta Gen 3 ($449), its first camera-free Ray-Ban Meta Audio glasses ($349), an FDA-cleared hearing-enhancement feature, wider Ray-Ban Display availability, and brought its Muse personal agent to glasses, Mac and a new pocket device.

- Keynote 2026-09-23 at Meta HQ, Menlo Park; event ran Sept 23-24
- Meta VR Glasses (Project Phoenix): ~100 g, about 5x lighter than Quest 3; 5K micro-OLED display; tethered compute puck; eye + hand tracking, no controllers; $1,299.99; ships spring 2027
- Ray-Ban Meta (Gen 3): $449; slimmer, action button, longest battery life (price per VR.org)
- Ray-Ban Meta Audio: first camera-free Meta glasses, $349, 12-hour battery (price per VR.org)
- Hearing enhancement on glasses, FDA-cleared: $149.99 or included in Meta One subscription (US, later 2026)
- Ray-Ban Display now in Canada and UK; France, Italy, Germany from Oct 13
- Muse agent: realtime voice, Muse Realtime Avatar, glasses support, Mac app with computer use, 'Muse Charm' pocket device
- Muse Realtime Avatar (Meta research blog 2026-09-23): Diffusion Transformer driven by Muse Realtime Voice speech tokens; 448x768 at 25 fps; ~870 ms from end of user turn to first response; 120-step teacher distilled to 2 steps (60x fewer evaluations); 12 concurrent sessions per GB200; preferred 78% vs Runway Characters and 88% vs HeyGen LiveAvatar in Meta's human tests; Meta Video Seal watermark; 18+ only, 'coming soon'
- The voice/avatar stack is led by Alexis Conneau, co-founder of WaveForms AI (acquired by Meta Aug 2025; ex-OpenAI GPT-4o voice)
- Over 100 glasses styles by year-end; new markets Singapore, South Korea, Mexico
- Muse Charm (Bloomberg): palm-sized, Tamagotchi-like device with a ~2-inch OLED screen, front and rear cameras and 5G, on sale by year's end
- Privacy: Meta will bring Private Processing to Ray-Ban Meta glasses (Wired) and let glasses users opt out of having their 'visual data' used for training or shown to contractors outside the US (Engadget)
- Horizon Create (mobile) and Horizon Studio (web) let people build games from AI prompts, playable on Facebook, Instagram and Horizon (The Verge, Sept 24)

#### What happened
Zuckerberg's Connect 2026 keynote centered on AI wearables and the Muse agent. The headline device was **Meta VR
Glasses**, an ultralight two-part headset (glasses plus belt-clip puck) with a 5K micro-OLED display and hand/eye input,
launching spring 2027 at $1,299.99. The AI-glasses line got **Ray-Ban Meta Gen 3**, the camera-free **Ray-Ban Meta
Audio**, health features (FDA-cleared hearing enhancement, workouts, nutrition tracking), shopping/product
identification, landmark-based navigation and Dolby Atmos spatial capture. **Muse** was extended to glasses and a
new pocket-sized voice device, **Muse Charm** (specs/pricing later in 2026).

#### Why it matters
Meta is betting that glasses become the primary interface for an always-present AI agent; Connect 2026 tied the
MSL model work (Muse Spark, Muse agent) directly to its hardware roadmap.

Prices for Gen 3 and Audio come from VR.org; Meta's own recap page did not list them in the version read.

Sources: [Meta - Everything we announced at Meta Connect 2026](https://www.meta.com/blog/meta-connect-2026-everything-we-announced/); [Engadget - Everything announced at Meta Connect 2026](https://www.engadget.com/2267230/everything-announced-at-meta-connect-2026/); [VR.org - Meta Connect 2026: everything announced](https://vr.org/meta-connect-2026); [TechCrunch - Everything new coming to Meta's AI agent Muse](https://techcrunch.com/2026/09/23/everything-new-coming-to-metas-ai-agent-muse/); [Meta AI research blog - Bringing your Muse to life (Muse Realtime Avatar)](https://research.meta.ai/blog/bringing-your-muse-to-life); [Alexis Conneau on X - introducing Muse Realtime Avatar (2026-09-24)](https://x.com/alex_conneau/status/2103143665577423347); [Latent Space AINews - Meta Connect 2026: Muse glasses, voice, video and Charm](https://www.latent.space/p/ainews-meta-connect-2026-muse-glasses); [Meta Connect Keynote 2026 (YouTube, Meta)](https://www.youtube.com/watch?v=SdKFDIAGF24); [Bloomberg: Meta debuts a dedicated palm-sized Muse Charm device](https://www.bloomberg.com/news/articles/2026-09-23/meta-debuts-a-dedicated-palm-sized-muse-charm-device-to-use-ai-on-the-go); [The Verge: Meta Muse gets video chat, email addresses, Mac computer use](https://www.theverge.com/tech/999454/meta-muse-ai-agent-video-chat-connect-2026); [Wired: Meta promises its smart glasses are going to be private soon](https://www.wired.com/story/meta-pinky-promises-its-smart-glasses-are-going-to-be-private-soon/); [Engadget: Meta will stop training its AI on visual data from its smart glasses if you opt out](https://www.engadget.com/2267227/meta-will-stop-training-its-ai-on-visual-data-from-its-smart-glasses-if-you-opt-out/); [The Verge: Meta Horizon Create and Studio for AI-made games](https://www.theverge.com/games/999972/meta-horizon-create-studio-ai-games); [Meta Newsroom - The biggest news from Connect 2026](https://about.fb.com/news/2026/09/the-biggest-news-from-connect-2026/); [Meta Newsroom - Introducing Meta VR Glasses](https://about.fb.com/news/2026/09/introducing-meta-vr-glasses-3d-movies-immersive-live-sports-100-grams/); [Meta Newsroom - Introducing Ray-Ban Meta Audio and more AI glasses styles](https://about.fb.com/news/2026/09/introducing-ray-ban-meta-audio-glasses-new-styles-plus-muse/); [Meta Newsroom - New features for Meta Ray-Ban Display](https://about.fb.com/news/2026/09/new-features-for-meta-ray-ban-display-navigation-hologram/)

### 2026-09-23: ChatGPT Voice gets plugins and moves into ChatGPT Work: spoken requests can now drive agent tasks

**; OpenAI; Products; confidence medium; https://postcutoff.com/e/2026-09-23-chatgpt-voice-plugins-work/

On 2026-09-23 OpenAI added plugin and connected-app support to ChatGPT's Live voice mode (GPT-Live-1 / mini) on web, iOS and Android, and put Voice inside ChatGPT Work. Users can now ask by voice for documents, slides, spreadsheets, connected-app actions or browser tasks. Consequential actions still need an on-screen approval; spoken approval is not accepted.

- Release-notes title (per press): 'Use plugins in Voice and get work done by speaking'
- Live voice + plugins: web, iOS, Android; Free and Go get the plugins their plan supports
- Voice in Work: web, mobile and desktop; needs both Voice and Work access; Plus and Pro get a Work tab in the mobile app (press)
- Approvals only through on-screen controls ('spoken approval is not supported'); one Voice conversation per account at a time
- Unfinished voice tasks can be continued in text; Work tasks started by voice count against Work usage
- Voice limits (Unite.AI): Go 3 h GPT-Live-1 mini, Plus 3 h GPT-Live-1, Pro $100 15 h, Pro $200 unlimited; Enterprise/Edu 1.25 credits/min or $0.05/min

#### What happened
OpenAI connected its full-duplex voice models (GPT-Live) to the same plugins and connected apps that text ChatGPT uses,
and made Voice an input to ChatGPT Work, its agent workspace for documents, slides, spreadsheets and browser tasks.
Reasoning-heavy parts are handed to text models (press mentions GPT-5.6 / GPT-6 Astra) while the conversation continues.

#### Why it matters
Voice stopped being a chat-only mode in the largest consumer assistant and became a way to start agent work. OpenAI's
rule that approvals must be tapped, not spoken, is an early safety convention for voice agents.

Confidence is medium because OpenAI's release notes returned 403 to our tools. The facts come from press summaries that
quote them.

Sources: [OpenAI Help Center - ChatGPT release notes (2026-09-23 item; 403 to our fetcher)](https://help.openai.com/en/articles/6825453-chatgpt-release-notes); [Unite.AI - OpenAI brings plugins to Live voice and Voice to Work in ChatGPT](https://www.unite.ai/openai-brings-plugins-to-live-voice-and-voice-to-work-in-chatgpt/); [AI Weekly - OpenAI wires ChatGPT Voice into Work agent and GPT-6 models](https://aiweekly.co/alerts/openai-wires-chatgpt-voice-into-work-agent-and-gpt-6-models); [Chat GPT AI Hub - ChatGPT Voice adds plugins and Work tasks (approvals, text handoff, data boundaries)](https://chatgptaihub.com/chatgpt-voice-plugins-work-connected-apps-on-screen-approvals-text-handoff-data-boundaries); [OpenAI on X: ChatGPT Voice can now use plugins and GPT-6 models](https://x.com/OpenAI/status/2102808325742322002)

### 2026-09-23: Claude Code silently ignored AGENTS.md when telemetry was off; fixed in v2.1.281

**; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-09-23-claude-code-agents-md-telemetry-bug/

On Sept 23, 2026 a developer blog (blog.szypowi.cz) showed that Claude Code (v2.1.277–2.1.280) loaded AGENTS.md instruction files only when telemetry was enabled: the loader sat behind a remote feature flag (tengu_agents_md_mod) that defaulted to off, so users who disabled telemetry or nonessential traffic had their AGENTS.md silently dropped. An Anthropic engineer said on Hacker News the same day that v2.1.281 fixed it.

- Affected: Claude Code 2.1.277 (measured also on 2.1.280); with telemetry or nonessential traffic disabled, 'with either variable set, AGENTS.md never loaded', with no warning
- Cause: AGENTS.md support gated by the remote feature flag tengu_agents_md_mod, which defaults to false
- Found with a canary word in a test AGENTS.md plus `claude -p` runs, then confirmed in the binary; GitHub issue #95690
- Workaround: a one-line CLAUDE.md containing `@AGENTS.md`
- Fix: 'it's already been fixed as part of v2.1.281 releasing today' (Anthropic's mpoteat on HN)
- Author: 'A privacy setting should never quietly switch off unrelated local behavior.'

#### What happened
AGENTS.md is the cross-tool convention for project instructions for coding agents (this repository uses one). Claude Code had added support for it,
but only behind a server-side flag. Privacy-minded users who had turned off telemetry never got the flag and never got their instructions loaded.

#### Why it matters
A small bug, but a clear example of how remote feature flags can make an agent's behavior depend on unrelated privacy settings. Users could not tell
whether their instructions were being read. It was fixed within a day of going viral.

Sources: [Claude Code reads AGENTS.md only when telemetry is on](https://blog.szypowi.cz/p/claude-code-reads-agents.md-only-when-telemetry-is-on/); [Hacker News discussion (485 points) incl. Anthropic fix confirmation](https://news.ycombinator.com/item?id=49814947)

### 2026-09-23: Google adds encrypted, persistent cross-device memory to Private AI Compute

**; Google, Google DeepMind; Products; confidence high; https://postcutoff.com/e/2026-09-23-google-private-ai-compute-memory/

On Sept 23, 2026 Google DeepMind described an update to Google's Private AI Compute platform. Cloud AI can now keep long-term, cross-device memory for a user in per-user encrypted storage whose keys stay only on the user's own devices, so that, Google says, the data is "inaccessible to anyone else, even Google". Data is decrypted only briefly inside hardware-isolated secure enclaves. Google also published a tamper-proof public record of its server software and the results of an independent security audit.

- Until now Private AI Compute (and similar industry systems) was 'stateless', wiping all context when a task ended; the new layer adds persistent memory
- Design: an end-to-end encrypted channel from device to an attested enclave; the enclave temporarily decrypts data in isolated memory, saves new context and re-encrypts it; per-user databases are protected by device-derived keys
- Transparency: updated technical whitepaper, a tamper-proof public record (log) of server software that devices can verify before sending data, and an independent audit by 'a leading cybersecurity firm' (not named in the post)
- Example uses: continuing on a laptop what was seen through smart glasses; resuming conversations across mobile and web
- Co-developed by Google DeepMind with Platforms & Devices, Core and Cloud teams

#### What happened
Google extended Private AI Compute, its secure-enclave cloud platform for processing personal data with Gemini, so that it can **remember**.
A user's context is stored in a per-user encrypted "vault" in the cloud. The keys stay on the user's devices, and data is decrypted only
inside attested, hardware-isolated enclaves while a request is handled. Google says this solves a long-standing trade-off: frontier models
need cloud compute, but strong privacy has usually meant on-device processing, and cloud enclaves could not keep state between requests.

#### Why it matters
Personal agents that remember users across phones, laptops and glasses need long-term memory. Google is betting that cryptographic
guarantees, rather than policy promises, will make users willing to let the cloud hold that memory. Other enclave-based systems in the industry (such as Apple's Private
Cloud Compute) are stateless, so this adds persistence. It also comes as Gemini replaces Google Assistant on Android.

Sources: [Google DeepMind: Advancing Private AI Compute with secure, server-side memory](https://deepmind.google/blog/advancing-private-ai-compute-with-secure-server-side-memory/)

### 2026-09-23: Made on YouTube 2026: conversational AI video editing, prompt-built custom feeds, live auto-dubbing and voice likeness detection

**; YouTube, Google; Products; confidence high; https://postcutoff.com/e/2026-09-23-made-on-youtube-2026-ai-features/

At Made on YouTube on Sept 23, 2026, YouTube announced 30+ updates, most of them AI features. They include a conversational editing tool in Shorts and the YouTube Create app, homepage "custom feeds" built from natural-language prompts, real-time auto-dubbing for live streams, Ask YouTube as a shopping guide, likeness detection on mobile with speaking-voice matching, and conversational Ask Music and AI "Podcast Lineup" previews in YouTube Music.

- Conversational editing in Shorts and YouTube Create: suggested or custom prompts produce a first draft, trim pauses and reorder for pacing (YouTube Blog)
- Custom Feeds: viewers design and save homepage feeds with prompts, e.g. 'Witty podcasts for my daily 30-minute commute' (coming 'soon')
- Live: real-time auto-dubbing, 'Live showdowns', 'Watch With' expanded to long-form
- Likeness detection moves to mobile and adds speaking-voice detection alongside face detection, to find AI-altered content using a creator's identity
- YouTube Studio: A/B testing up to three cuts of a video; Ask Studio drafts responses to brand briefs
- YouTube Music: Ask Music (conversational queue building) and Podcast Lineup (weekly AI-voiced previews of recommended shows)

#### What happened
YouTube's annual creator event turned generative and conversational AI into default editing, discovery and dubbing tools for its creators and
viewers.

#### Why it matters
YouTube is the largest video platform. Moving recommendations from opaque algorithms to prompt-defined feeds, and adding voice-likeness
detection, are notable steps in how AI reshapes media distribution and identity protection.

Sources: [YouTube Blog: Made On YouTube 2026, all announcements](https://blog.youtube/news-and-events/innovation-youtube-era-made-on-viewers-creators/); [YouTube Blog: Creator tools 2026 (Gemini AI and Studio updates)](https://blog.youtube/news-and-events/made-on-youtube-new-tools-power-creation-journey/); [YouTube Blog: Ask Music and Podcast Lineup](https://blog.youtube/news-and-events/made-on-youtube-music-new-discovery-features/); [Tom's Guide: YouTube reveals new AI capabilities](https://tomsguide.com/ai/youtube-reveals-new-ai-capabilities-heres-how-they-work)

### 2026-09-25: Microsoft unveils the 'new Copilot' with Home, Code and Autopilot agents, offering Astra and Fable models

****; Microsoft; Products; confidence high; https://postcutoff.com/e/2026-09-25-microsoft-new-copilot-home-code-autopilot/

On Sept 25, 2026 Microsoft introduced the "new Copilot", which Nadella called "a new OS for work". Home merges Chat and Cowork and embeds Word, Excel and PowerPoint. Code builds apps and dashboards on GitHub Copilot technology, running in a new Copilot Managed Runtime. Autopilot (formerly Scout) is a persistent named agent with a role and avatar. Users can choose OpenAI's Astra, Anthropic's Fable or an Auto router.

- Announced Sept 25, 2026 (blog by Jared Spataro)
- Home = Chat + Cowork + 'Office in Copilot' (Word, Excel, PowerPoint inside Copilot)
- Code: app/dashboard builder on GitHub Copilot tech; apps run on the Microsoft Copilot Managed Runtime (preview)
- Autopilot (previously 'Scout'): persistent agent with name, role and avatar; private preview from end of month
- Model choice: frontier models Astra (OpenAI) and Fable (Anthropic), or 'Auto'
- Billing: fixed per-user license for everyday use plus usage-based billing for Cowork, Code, Autopilot and frontier models
- Autopilot (formerly Scout) is built on OpenClaw: Microsoft's Omar Shahine said his team works with Peter Steinberger and the OpenClaw Foundation, and Steinberger said Microsoft has worked with the project since March
- Nadella (per The Verge interview): Autopilot 'should also go to the consumer side', i.e. a Muse competitor

#### What happened
Microsoft rebuilt Copilot around three surfaces (a workspace, an app builder and an always-on agent) and put both OpenAI's and Anthropic's top models inside it, with usage-based pricing for heavy agent work.

#### Why it matters
Microsoft is moving from per-seat assistant pricing to metered agents, and treats frontier models from rival labs as interchangeable components.

Sources: [Sumit Chauhan (Microsoft) on X: the full power of Office inside the Copilot app](https://x.com/sumit_c/status/2103486852833620405); [Bloomberg: Microsoft abandons personal AI chatbot race with Copilot reboot](https://www.bloomberg.com/news/articles/2026-09-25/microsoft-abandons-personal-ai-chatbot-race-with-copilot-reboot); [Microsoft: Introducing the new Copilot with Home, Code and Autopilot](https://blogs.microsoft.com/blog/2026/09/25/introducing-the-new-copilot-with-home-code-and-autopilot/); [Microsoft Source EMEA: New Microsoft Copilot brings Home, Code and Autopilot together](https://news.microsoft.com/source/emea/2026/09/new-microsoft-copilot-brings-home-code-and-autopilot-together/); [Gizmodo: Microsoft thinks it's finally figured out Copilot](https://gizmodo.com/microsoft-thinks-its-finally-figured-out-copilot-this-time-2000817460); [The new Copilot (official video)](https://www.youtube.com/watch?v=OgInADh5Tcs); [The Verge: Microsoft launches Copilot super app, rebrands Scout as Autopilot](https://www.theverge.com/news/1000532/microsoft-copilot-super-app-chat-coding-autopilot); [Jacob Andreou (X Article): The New Copilot](https://x.com/jacobandreou/status/2103470116906176622); [Omar Shahine on X: Autopilot is built on OpenClaw](https://x.com/OmarShahine/status/2103480227561079264); [Peter Steinberger on X: Microsoft shipped on top of OpenClaw](https://x.com/steipete/status/2103491173927272531)

### 2026-09-28: Meta launches an Enterprise Platform division led by ex-MongoDB CEO CJ Desai, and Muse for Small Business

****; Meta; Products; confidence high; https://postcutoff.com/e/2026-09-28-meta-enterprise-platform-muse-business/

On Sept 28, 2026 Meta created the Meta Enterprise Platform, led by former MongoDB CEO Chirantan "CJ" Desai as Chief Enterprise Platform Officer reporting to Zuckerberg, with Muse, Meta Business Agent, the Muse API and Muse Code as launch products. On Sept 29 it launched Muse for Small Business in the US and Canada: free for most features, 15+ connectors, and approval required before publishing, sending or spending.

- Sept 28: Meta Enterprise Platform division; CJ Desai (ex-MongoDB CEO) reports to Zuckerberg
- Initial products: Muse, Meta Business Agent, Muse API, Muse Code; no pricing or model ids disclosed
- Sept 29: Muse for Small Business in the US and Canada, free for most features with paid upgrades
- 15+ connectors incl. Asana, Box, Canva, Dropbox, Figma, Notion, QuickBooks, Shopify, Slack, Stripe, Zoom, plus Facebook/Instagram business accounts; custom connectors supported
- Asks for approval before publishing, sending or spending
- Consumer Muse momentum: #1 on the US App Store (Sept 18); 2.3M–4.3M downloads by ~Sept 24 depending on the analytics firm (TechCrunch)

#### What happened
After its consumer agent Muse took off, Meta set up a dedicated enterprise business and pushed Muse to small businesses, which already run on its ad and messaging platforms.

#### Why it matters
Meta is now competing directly with Microsoft, Google, OpenAI and Anthropic for enterprise agent spending, not only consumer attention.

Sources: [Meta: Launching Meta Enterprise Platform](https://about.fb.com/news/2026/09/launching-meta-enterprise-platform/); [Meta: Introducing Muse for Small Business](https://about.fb.com/news/2026/09/introducing-muse-small-business/); [TechCrunch: Meta launches enterprise AI platform, hires MongoDB CEO](https://techcrunch.com/2026/09/28/meta-launches-enterprise-ai-platform-hires-mongodb-ceo-to-lead-new-initiative/); [TechCrunch: Meta is expanding Muse to small businesses](https://techcrunch.com/2026/09/29/meta-is-expanding-its-ai-agent-muse-to-small-businesses/); [CNBC: Meta launches Muse for Small Business](https://www.cnbc.com/2026/09/29/meta-launches-muse-for-small-business-zuckerberg-pushes-enterprise-ai.html); [TechCrunch: Meta is putting its muscle behind Muse as the AI app takes off](https://techcrunch.com/2026/09/25/meta-is-putting-its-muscle-behind-muse-as-the-ai-app-takes-off/); [WSJ: Meta seeks payoff from AI spending with new push for business customers](https://www.wsj.com/tech/ai/meta-seeks-payoff-from-ai-spending-with-new-push-for-business-customers-8b9ca5bc)

### 2026-09-28: Gemini fully replaces Google Assistant on Android; Gemini gets 'Call for Me', and Gems become skills

***; Google; Products; confidence high; https://postcutoff.com/e/2026-09-28-gemini-replaces-google-assistant-android/

In the week of Sept 24–28, 2026 Google removed the option to switch back to Google Assistant on Android phones, tablets, Wear OS, headphones and Android Auto, making Gemini the only assistant. It also began testing 'Call for Me', in which Gemini phones businesses for the user (navigating menus, waiting on hold, showing a live transcript). And it announced that Gems will be migrated into slash-invoked 'skills' from Nov 17.

- Sept 28: 'Switch to Google Assistant' removed on phones, tablets, Wear OS, headphones and Android Auto; Nest/Home speakers unaffected
- Sept 24: 'Call for Me' test: Gemini calls businesses from the user's number, navigates phone menus, waits on hold, shows a live transcript; US Pixel 11, paid Gemini plan, Phone app beta
- Sept 27/28: Gems become 'skills' invoked with '/', automatic migration from 2026-11-17
- Also that week: new Connected Apps (Sept 23), Google Wallet integration (Sept 28), Flipkart purchases test in India (Sept 26)

#### What happened
Google completed the switch from Assistant to Gemini on Android and gave it more real-world agent abilities, such as making phone calls.

#### Why it matters
Google Assistant, the default assistant on billions of devices since 2016, ends on Android, replaced by an LLM agent.

Sources: [9to5Google: Google Assistant is gone from Android, Gemini takes over](https://9to5google.com/2026/09/28/google-assistant-gemini-android/); [TechCrunch: Google tests letting Gemini make phone calls](https://techcrunch.com/2026/09/24/google-tests-letting-gemini-make-phone-calls-initially-for-us-pixel-owners/); [TechCrunch: Google is killing off Gemini's Gems in favor of skills](https://techcrunch.com/2026/09/28/google-is-killing-off-geminis-gems-in-favor-of-skills/); [Google: New Connected Apps in Gemini](https://blog.google/innovation-and-ai/products/gemini-app/new-connected-apps-gemini/); [9to5Google: Gemini Gems are becoming skills](https://9to5google.com/2026/09/27/gemini-gems-skills/)

### 2026-09-29: OpenAI DevDay 2026: dots agents, GPT-6.1 Sol, Ultrafast, a $500 Pro plan and 20+ launches

****; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-09-29-openai-devday-2026/

At DevDay 2026 (Fort Mason, San Francisco, Sept 29, 2026) OpenAI announced more than 20 launches. The headline items were "dots", always-on personal agents powered by GPT-6 Astra; GPT-6.1 Sol, which OpenAI says nearly matches Astra at one-fifth of its price; an "Ultrafast" speed tier (up to 8x faster in Codex, 6x in the API, at 6x the price); and a new $500/month "Pro 500" ChatGPT plan, with the $200 plan's allowance halved. OpenAI also added ChatGPT Space/Pages, plugin extensions, a Decisions API, computer use in the Agents API, Codex Security Cloud, "Sign in with ChatGPT" plan sharing and an enterprise Marketplace. The event came a day after OpenAI cancelled GPT-6.1 Astra over failed alignment tests.

- Keynote by Sam Altman at 10:00 PT, Sept 29, 2026 at Fort Mason, San Francisco; livestreamed on openai.com/live and YouTube; OpenAI's recap lists 'more than 20 major announcements' and cites 1.2B weekly users
- dots: always-on agents powered by GPT-6 Astra, each with its own cloud computer and browser, 4,000+ connectable apps, reachable in ChatGPT, Slack and Teams (texting 'coming soon'); rolling out to Pro (incl. Pro 100) and Business Premium in eligible markets, Enterprise/Edu/Healthcare beta when admins enable it; the first dot is included in the plan and dot conversations don't count toward usage limits
- GPT-6.1 Sol (API id gpt-6.1-sol): $2 input / $0.10 cached input / $10 output per 1M tokens; 'near-Astra intelligence' at one-fifth of Astra's standard prices; in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu (not yet in Chat); also GA in GitHub Copilot the same day
- Ultrafast service tier: up to 8x faster token generation (300 tokens/s) in Codex and up to 6x in the API; GPT-6 Astra Ultrafast costs $60 input / $6 cached / $300 output per 1M tokens (6x standard); API param service_tier: "ultrafast"; GPT-6.1 Sol Ultrafast 'coming soon'
- Pro 500 plan: $500/month, 25x the Plus allowance, includes Ultrafast (Pro 500 and Enterprise only in ChatGPT Work/Codex); Pro 200 reopened to new subscribers but cut from 20x to 10x Plus and from 200 to 100 GPT-6 Pro messages/week; existing subscribers keep old limits through Oct 29, 2026
- Private Intelligence: Zero Data Retention with Private Safety Processing (automated safety reviews without OpenAI staff seeing content) available now; a preview of Private Inference (confidential computing) is due 'this fall'; Altman named Cisco, Databricks and Snowflake as design partners
- Codex: cloud Codex with reusable environments (Plus and above); refreshed Codex CLI with voice control and an /agents view (all plans); code review in the ChatGPT desktop app with automatic cloud reviews (all plans); Codex Security Cloud scans whole GitHub repos and includes Daybreak Blue cyber models without a separate Daybreak application (Pro, Business, Enterprise, Edu)
- API: Decisions API (a version of GPT-6 Luna answering user-defined questions with fixed answer sets, ~150 ms vs 1.6 s for plain Luna per OpenAI's chart; limited preview, broad release 'in the coming days', price not announced); Agents API (public beta since Sept 10) gains computer use, multi-agent, tool search and compaction; Bedrock Managed Agents on AWS now built on the Agents API core
- ChatGPT for teams: ChatGPT Space (shared team hub with dots), Pages (collaborative human/agent documents), collaborative Slides ('coming weeks'), Teams and team tasks, @ChatGPT in Slack and Microsoft Teams, a Meetings plugin (macOS beta; audio deleted after notes), shareable profiles
- Plugins: plugin extensions (sidebar homes, interactive panels, custom file viewers) on all plans, Plugin Creator and a new submission flow, plugins inside Sites, and support for the proposed MCP Events specification for event-triggered automations
- Sign in with ChatGPT: Plus and Pro users can spend their plan allowance in 16 partner tools (e.g. Devin, Notion, Vercel, T3, OpenClaw, Dactyl, Amp, Warp; Lovable 'coming soon') with per-app caps
- OpenAI Marketplace: eligible enterprises can spend part of their OpenAI commitment on 32 partners' software (Adobe, Figma, Salesforce, ServiceNow, HubSpot, Sierra, Decagon, Harvey, Legora, CrowdStrike, Palo Alto Networks, Baseten, ...)
- Not announced (despite pre-event reports): no 'GPT-6 Cyber' model and no hardware device; the rumoured always-on agent ('o'/'Aeon') shipped under the name dots, and the rumoured $500 'Pro Max' plan shipped as 'Pro 500'
- Context: GPT-6.1 Astra was cancelled on Sept 28 over failed alignment tests; protesters gathered outside the venue; President Greg Brockman skipped DevDay to attend a White House AI lunch
- Baseten (Sept 29) is one of the first open-model inference providers in the OpenAI Marketplace: enterprises can spend existing OpenAI commitments on open models served by Baseten inside Codex or via the Responses API
- ChatGPT Space replaces the ChatGPT Library for eligible Pro, Business and Enterprise accounts (Projects stay separate); its main format is Pages, editable by teammates, ChatGPT and dots (OpenAI Help Center)
- Follow-ups: Sites became generally available for ChatGPT Enterprise on Oct 15, 2026, with US and EU data residency for new Sites in eligible workspaces, team apps with databases and authentication, and Plugins inside Sites (@OpenAIDevs, ~63k views). The plugin-extension spec is published as the openai/mcp-extensions GitHub repo (created Sept 29; ~1,075 stars by Oct 16): sidebar entrypoints, custom file-viewer handlers and composer @-mentions on top of MCP
- Follow-up (Oct 12, 2026): OpenAI's product film 'Code review is no longer the bottleneck' (~88k YouTube views by Oct 18) shows Codex Code Review as a PR inbox with automatic cloud reviews, local private reviews, custom review instructions and PR chat on GPT-6.1 Sol; its description says GitHub code review is generally available and GitLab support is in preview (OpenAI's help article on reviewing pull requests also lists GitLab as preview-only, without cloud reviews)

#### What happened
OpenAI held DevDay 2026 on Tuesday, Sept 29, 2026 at Fort Mason in San Francisco. Sam Altman gave the keynote at 10:00 PT. Beforehand
he told CNBC there would be "more than 20 announcements", and on Sept 28 he had posted that OpenAI had "found a new thing". OpenAI's recap
post groups the launches into five areas:

**New ways of working**
- **dots**: always-on agents powered by GPT-6 Astra, each with its own cloud computer. See `2026-09-29-openai-dots`.
- **GPT-6.1 Sol**: an upgrade to GPT-6 Sol, released a week after it. OpenAI says it gives "near-Astra intelligence" at one-fifth of Astra's
  standard token prices ($2/$10 per 1M; cached input $0.10). See `2026-09-29-gpt-6-1-sol`.
- **Ultrafast**: a premium speed tier. It is up to 8x faster (300 tokens/s) in Codex and up to 6x faster in the API, at 6x the standard API
  price. GPT-6 Astra Ultrafast is out now; GPT-6.1 Sol Ultrafast is "coming soon". See `2026-09-29-chatgpt-pro-500-ultrafast`.
- **OpenAI Private Intelligence**: Zero Data Retention with Private Safety Processing lets automated safety reviews run without OpenAI
  staff seeing the content. A preview of **Private Inference**, which combines confidential computing with verifiable controls, is due
  "this fall". Altman said OpenAI designed it with Cisco, Databricks and Snowflake.

**Codex and the API**
- **Codex in the cloud** with reusable, shareable development environments (Plus, Pro, Business, Healthcare, Edu, Enterprise).
- **Refreshed Codex CLI**: voice control, an `/agents` view for tracking several tasks, better worktree and session handling (all plans).
- **Code review** in the ChatGPT desktop app, feeding GitHub PRs and GitLab MRs, with automatic cloud reviews (all plans).
- **Codex Security Cloud**: on-demand or scheduled scans of whole GitHub repositories and new commits. Codex deduplicates findings and
  prepares fixes. It includes Daybreak Blue models without a separate Daybreak application (Pro, Business, Enterprise, Edu).
- **Decisions API**: "real-time decision-making" that focuses a version of Luna on user-defined questions with a fixed set of answers.
  Input can be text or images. OpenAI's chart shows about 150 ms per decision versus 1.6 s for GPT-6 Luna through the regular API. It is
  in limited preview, with broad release "in the coming days". No price has been published. Press coverage (The New Stack, The Decoder)
  presented it as a response to TypeSafe's "Jev" decision model.
- **Agents API with computer use**: the Agents API, a managed Codex harness in public beta since Sept 10, now supports computer use
  in an OpenAI-hosted browser. It also gains Codex's multi-agent features, tool search and context compaction.
- **Bedrock Managed Agents, powered by OpenAI**: now built on the Agents API core, so OpenAI agents can run entirely in AWS.

**Plugins**
- **Plugin extensions** give a plugin a sidebar home, interactive panels and file viewers (all plans). Also new: Plugin Creator, a
  redesigned submission flow and better ranking, plugins inside **Sites** (Business, Enterprise, Healthcare, Edu), and support for the
  proposed **MCP Events** specification, which lets connected-app events trigger automations.

**People and AI working together**
- **ChatGPT Space**: a shared hub for teammates, ChatGPT and dots (Pro, Business, Enterprise; desktop and web).
- **Pages**: collaborative documents for humans and agents.
- **Collaborative slides** ("coming weeks"), exportable to PowerPoint and Google Slides.
- **Teams and team tasks** for scheduled or event-triggered recurring work.
- **@ChatGPT in Slack and Microsoft Teams**, usable without an individual license.
- **Meetings plugin**: notes and action items. It is a macOS beta for Pro and Business, and audio is deleted after notes are made.
- **Shareable profiles** for showcasing Sites and plugins.

**Subscriptions**
- **Sign in with ChatGPT**: Plus and Pro users can spend their plan allowance in 16 partner tools, including Devin, Notion, Vercel, T3,
  OpenClaw, Dactyl, Amp and Warp, with a weekly cap for each app. Identity sign-in is available globally.
- **Pro 500**: a $500/month plan with 25x the Plus allowance and Ultrafast. At the same time, **Pro 200** reopened to new sign-ups with
  half its former allowance (10x Plus instead of 20x).
- **OpenAI Marketplace**: enterprises can spend part of their OpenAI commitment on software from 32 partners.

### Around the keynote (CNBC interview with Altman)
- GPT-6.1 Astra's cancellation was "normal course": "Often we build a model, we test it, it doesn't meet our standards, we change it, we
  launch it later." OpenAI has "many great new models to come."
- Hardware: "something that's worth waiting for", but "I am not worried about being first with new hardware." No device was shown.
- IPO: "I don't have a particular timeline in mind", and "this is a time to put safety and mission first."
- He called Nvidia's new agent safety platform "a good thing" but "not a full solution", and said Meta's Muse "seems like a nice product."
- Protesters rallied outside the venue. Greg Brockman skipped DevDay to attend the White House AI lunch (`2026-09-29-white-house-ai-summit`).

### Pre-event rumours vs. reality
- The rumoured always-on agent ("o", reportedly code-named "Aeon") launched as **dots**.
- The rumoured $500/month "ChatGPT Pro Max" launched as **Pro 500**.
- **GPT-6 Cyber** was *not* announced. Cyber work was served through Daybreak Blue access inside Codex Security Cloud.
- **No hardware device** was announced.

#### Why it matters
DevDay marks OpenAI's shift from chat and coding tools toward persistent, proactive agents (dots), and toward ChatGPT as a workspace
platform (Space, Pages, plugin extensions, Marketplace). It is a direct answer to Meta's Muse. Pricing moved in two directions at once.
Model prices kept falling: GPT-6.1 Sol offers near-flagship quality at $2/$10, with cached input at $0.10. Subscriptions were re-tiered
toward pay-per-use: the $200 plan's allowance was halved, a $500 tier was added, and paid speed became a product (Ultrafast at 6x the
price). All of this came one day after OpenAI withheld GPT-6.1 Astra on safety grounds. That put the safeguards for always-on agents
running on GPT-6 Astra under close scrutiny.

Sources: [TechCrunch: OpenAI's latest features take direct aim at the App Store model](https://techcrunch.com/2026/09/29/openais-latest-features-take-direct-aim-at-the-app-store-model/); [OpenAI: DevDay 2026 Recap](https://openai.com/index/devday-2026-recap/); [OpenAI: Introducing dots](https://openai.com/index/introducing-dots/); [OpenAI: Introducing GPT-6.1 Sol](https://openai.com/index/introducing-gpt-6-1-sol/); [OpenAI: How we build safety, security, and privacy into dots](https://openai.com/index/how-we-build-safety-security-and-privacy-into-dots/); [OpenAI API docs: Ultrafast mode](https://developers.openai.com/api/docs/guides/ultrafast-mode); [OpenAI API docs: Private Safety Processing](https://developers.openai.com/api/docs/guides/private-safety-processing); [OpenAI API docs: Agents API computer use](https://developers.openai.com/api/docs/guides/agents-api/tools/computer-use); [OpenAI API pricing (incl. Ultrafast table)](https://developers.openai.com/api/docs/pricing); [OpenAI Help: About ChatGPT Pro tiers](https://help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers); [OpenAI Developers: Plugin extensions](https://developers.openai.com/plugins/build/extensions); [OpenAI Developers: MCP events](https://developers.openai.com/plugins/build/mcp-events); [OpenAI Marketplace](https://openai.com/business/marketplace/); [AWS: Bedrock Managed Agents powered by OpenAI](https://aws.amazon.com/bedrock/managed-agents-openai/); [CNBC live blog: OpenAI DevDay 2026](https://www.cnbc.com/2026/09/29/openai-devday-2026-live-updates.html); [The Verge: OpenAI DevDay 2026, the biggest news and announcements](https://www.theverge.com/ai-artificial-intelligence/1001681/openai-devday-2026-biggest-news-announcements); [The Decoder: OpenAI expands Codex and its API with security scans, a Decisions API and Ultrafast](https://the-decoder.com/openai-expands-codex-and-its-api-at-devday-with-security-scans-a-decisions-api-and-ultrafast/); [The Decoder: A new ChatGPT that looks more like an operating system](https://the-decoder.com/openais-reveals-a-new-chatgpt-that-looks-less-like-a-chatbot-and-more-like-an-operating-system/); [TechCrunch: OpenAI gives Codex reusable cloud environments](https://techcrunch.com/2026/09/29/openai-gives-codex-reusable-cloud-environments-that-work-across-devices/); [TechCrunch: OpenAI expands ChatGPT's plugins with app-like interfaces and automations](https://techcrunch.com/2026/09/29/openai-expands-chatgpts-plugins-with-app-like-interfaces-and-automations/); [The New Stack: 'Sign in with ChatGPT' lets subscriptions power third-party tools](https://thenewstack.io/sign-in-with-chatgpt/); [The New Stack: OpenAI's Decisions API built on Luna](https://thenewstack.io/openai-decision-api-luna/); [CNET: Everything announced at OpenAI DevDay](https://www.cnet.com/tech/services-and-software/everything-announced-at-openai-devday-subscription-changes-new-models-and-dots/); [Engadget live blog: OpenAI Dev Day 2026](https://www.engadget.com/2271985/openai-dev-day-live-blog-chatgpt-news/); [9to5Mac: OpenAI makes 20+ announcements at DevDay](https://9to5mac.com/2026/09/29/openai-teases-20-announcements-at-devday-watch-live/); [The Verge: Protesters gather at OpenAI's DevDay](https://www.theverge.com/ai-artificial-intelligence/1002201/openai-sam-altman-openai-devday-protests-ice-data-centers); [Sam Altman on X: 'We have found a new thing.' (Sept 28)](https://x.com/sama/status/2104661956879913457); [OpenAI on X: 'Get ready.' (20+ launches teaser)](https://x.com/OpenAI/status/2104651136699609518); [OpenAI on X: '10am PT, on the dot.'](https://x.com/OpenAI/status/2104934336206430527); [OpenAI on X: Codex Security Cloud](https://x.com/OpenAI/status/2104987422308335828); [Thibault Sottiaux on X: live posts on dots, 6.1 Sol, Decisions API, Codex Cloud](https://x.com/thsottiaux/status/2104987594719461796); [Simon Willison: OpenAI DevDay 2026 live blog](https://simonwillison.net/2026/Sep/29/openai-devday-2026-live-blog/); [Baseten: Baseten and OpenAI partnership](https://www.baseten.co/blog/baseten-openai-partnership/); [VentureBeat: OpenAI launches Dots and ChatGPT Space](https://venturebeat.com/technology/openai-launches-dots-always-on-ai-agent-coworkers-and-chatgpt-space-where-they-can-collaborate-with-human-teams); [Engadget: ChatGPT Space for work](https://www.engadget.com/2272248/chatgpt-space-for-work/); [OpenAI Help Center: ChatGPT Space sharing, data, and controls](https://help.openai.com/en/articles/20001544-chatgpt-space-sharing-data-and-controls); [OpenAI Help Center: Getting started with Space in ChatGPT](https://help.openai.com/en/articles/20001549-getting-started-with-space-in-chatgpt); [OpenAIDevs on X: Sites is now generally available for Enterprise (Oct 15)](https://x.com/OpenAIDevs/status/2110832979656524283); [GitHub: openai/mcp-extensions (ChatGPT-specific MCP extensions for plugins)](https://github.com/openai/mcp-extensions); [OpenAI (YouTube): Code review is no longer the bottleneck (Oct 12)](https://www.youtube.com/watch?v=IZXv6a78NDw); [OpenAI Help: Review pull requests with Codex (403 for us; status per search snippet)](https://help.openai.com/en/articles/20001552-review-pull-requests-with-codex)

### 2026-09-29: US launches America.gov, an AI chatbot portal to federal services powered by Gemini and Grok

***; White House, National Design Studio, Google, xAI; Products; confidence high; https://postcutoff.com/e/2026-09-29-america-gov-ai-portal/

On Sept 29, 2026 the Trump administration launched America.gov, an AI-powered search and chat portal that answers questions about federal services by drawing on about 29,000 federal websites. US Chief Design Officer Joe Gebbia (National Design Studio) said it runs on Google's Gemini and xAI's Grok. At launch it points users to the right services; completing forms and applications on the site is planned for 2027.

- Launched Sept 29, 2026, hours before the White House AI lunch; Trump: 'You'll now have one front door for every single question' (FedScoop)
- Built under Joe Gebbia, Airbnb co-founder and US Chief Design Officer, who leads the National Design Studio created in 2025 ('America by Design')
- Models: Google Gemini and xAI Grok (Gebbia to CNBC)
- Draws on roughly 29,000 federal websites; example tasks: replacing a Social Security card, registering to vote, comparing Medicare plans, registering a business, booking campsites, passports
- Planned update by early 2027: complete forms, applications and renewals on the site and track progress; identity via Login.gov (FedScoop)
- Privacy: Gebbia called it 'private by design'; FedScoop reports providers reportedly don't retain prompts, responses are cached up to two hours by prompt hash, and the White House gave no contract or procurement details
- Day one: the chatbot contradicted Trump's claims, saying official assessments found no 2020 election fraud 'at a scale that changed the outcome', and rejecting his 'worst inflation' claim with BLS data. Sen. Adam Schiff posted a screenshot: 'Finally, a straight answer from someone in the Trump administration' (AFP)
- By Wednesday Sept 30 it declined such questions: 'I do not provide political commentary or analysis of past elections.' White House: 'America.gov will continue to be updated to provide Americans the most accurate and reliable federal information', without saying whether settings were changed (AFP via TechXplore, RTE)
- Easter egg: asked to 'play Minecraft', the bot answered with a ~1,800-word bureaucratic rewrite of Minecraft's End Poem (by Julian Gough), with the love line turned into one about a ZIP code; TechCrunch's Amanda Silberling found it, it looked deliberate rather than a hallucination, and nobody has said who added it. By Oct 1 the bot replied that it 'can't play Minecraft or help with games' (Decrypt, The Star)

#### What happened
The White House opened a single AI "front door" to federal services, built by the National Design Studio on commercial models from Google and xAI.

#### Why it matters
It is one of the largest government deployments of a consumer-facing LLM assistant. It raises questions about accuracy, privacy and vendor
choice (xAI is owned by Musk, who attended the same day's AI lunch).

Sources: [Decrypt: US government's new AI chatbot has a weird Minecraft secret](https://decrypt.co/379739/america-gov-ai-chatbot-minecraft-end-poem); [The Star: Someone hid a Minecraft easter egg inside the White House's new AI chatbot (Oct 5)](https://www.thestar.com.my/tech/tech-news/2026/10/05/someone-hid-a-minecraft-easter-egg-inside-the-white-houses-new-ai-chatbot); [The Hill: Trump launches AI government website](https://thehill.com/homenews/administration/6117728-trump-launches-ai-government-website/); [FedScoop: Trump launches AI-fueled America.gov in bid to tie government services together](https://fedscoop.com/trump-launches-ai-site-america-gov/); [CNBC: New AI-powered government website uses Gemini, Grok, Trump official Gebbia says](https://www.cnbc.com/2026/09/29/trump-ai-gemini-grok.html); [Newsweek: What to know about America.gov](https://www.newsweek.com/trump-launches-america-gov-ai-government-website-12501100); [America.gov](https://www.america.gov); [AFP via TechXplore: US government's new AI chatbot pulls back from debunking Trump](https://techxplore.com/news/2026-10-ai-chatbot-debunking-trump.html); [RTE: US government's new AI chatbot stops debunking Trump](https://www.rte.ie/news/2026/1001/1593584-us-trump-ai-chatbot/); [Washington Post: We tested America.gov, Trump's new AI 'concierge'](https://www.washingtonpost.com/technology/2026/10/01/governments-new-chatbot-wont-talk-politics-it-will-fact-check-trump/); [Yahoo: Trump launches new AI tool and is immediately trolled](https://www.yahoo.com/news/politics/articles/trump-launches-ai-tool-immediately-191715754.html)

### 2026-09-30: Amazon plans 20,000+ AI smart glasses for delivery drivers by 2027, feeding its generative-AI mapping; privacy questions follow

**; Amazon; Products; confidence medium; https://postcutoff.com/e/2026-09-30-amazon-driver-smart-glasses/

Amazon says it will have more than 20,000 Smart Delivery Glasses in the field by the end of 2027 (Bloomberg: ~5,000 drivers in 2026). The glasses capture images that feed Amazon's generative-AI mapping platform. On Sept 30 a Bloomberg Opinion column raised privacy concerns over photos taken of homes and passers-by.

- Amazon (Ignite Live, Sept 21, 2026): 20,000+ Smart Delivery Glasses by end of 2027; new LED headlamp, hands-free scanning, proactive address verification; $1.9B more for Delivery Service Partners in 2027
- Amazon's 'Wellspring' generative-AI mapping has logged 202M parking spots, 2.8M building entrances and 85K mailrooms (Amazon X Article)
- Bloomberg: ~5,000 drivers wearing the glasses in 2026; the glasses capture several thousand static images per shift
- Asked about an opt-out for people photographed, Amazon's Chatterjee reportedly said 'We haven't thought about that' (Bloomberg, via X)

#### What happened
The glasses, introduced in 2025, display navigation and package details to drivers. Amazon says the new images improve its AI maps of
drop-off points.

#### Why it matters
It is one of the largest workplace deployments of AI smart glasses, and it turns delivery routes into a continuous image-collection network
for AI mapping.

Unverified: the "several thousand photos per shift" figure and the Chatterjee quote come via X posts summarizing Bloomberg.

Sources: [Bloomberg Opinion: Amazon drivers' smart glasses are going to be an image problem](https://www.bloomberg.com/opinion/articles/2026-09-30/amazon-drivers-smart-glasses-are-going-to-be-an-image-problem); [Amazon News on X (Sept 21): DSP investment and smart glasses](https://x.com/amazonnews/status/2102032303744491971)

### 2026-09-30: Claude for Government becomes generally available to US federal and state agencies in a FedRAMP High environment

**; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-09-30-claude-for-government-ga/

On Sept 30, 2026 Anthropic made Claude for Government generally available to US federal and state agencies after a public beta that began in July. It runs in a FedRAMP High authorized environment and has no seat fees: agencies prepay usage in fixed increments under a hard not-to-exceed cap. Claude Code CLI and Claude for Microsoft 365 entered early access under the same controls.

- GA Sept 30, 2026 for federal and state agencies; public beta since July
- FedRAMP High authorized environment; Anthropic's support FAQ says it runs in the Claude Desktop app connected to a dedicated government environment inside Palantir's FedRAMP High boundary
- Pricing: no seat fees; usage paid in fixed increments with a hard not-to-exceed cap; admins set spend limits and user tiers with model restrictions
- Features: desktop app with file access, Claude Code, skills, plugins, projects; audit logs, spend analytics, SSO and SCIM group mappings
- Early access: Claude Code CLI and Claude for Microsoft 365

#### What happened
Anthropic's government product moved from beta to general availability, with procurement shaped for agencies: capped prepaid
usage instead of per-seat licences.

#### Why it matters
It is a routine product step, but it widens frontier-model and coding-agent access across US government. It comes the same week
as the White House accord and the FTC probe of frontier labs.

Sources: [Claude blog: Claude for Government is now generally available](https://claude.com/blog/claude-for-government-is-now-generally-available); [Claude Help Center: Public sector FAQs](https://support.claude.com/en/articles/13756069-public-sector-faqs)

### 2026-09-30: Gemini rolls out reusable 'skills' to all users and will retire Gems (migration from Nov 17)

**; Google; Products; confidence high; https://postcutoff.com/e/2026-09-30-gemini-skills-replace-gems/

On Sept 30, 2026 Google rolled out skills in the Gemini app worldwide. Skills are reusable, stackable custom instructions invoked with a slash command inside any chat; they were previously exclusive to the paid Gemini Spark agent. They replace Gems: automatic migration of personal accounts' Gems begins Nov 17, 2026 (Workspace in March 2027).

- Invoke with '/skill-name' inside a chat; several skills can be combined in one session; skills can reference files such as PDFs and images
- Previously a Gemini Spark (paid) feature; now for personal Google accounts and Workspace users
- Gems to skills migration: personal accounts from Nov 17, 2026 (originally planned for Oct 20); Workspace March 2027; schools June 2027 (TechRepublic, Yahoo Tech)
- Exact Workspace dates (Google's Workspace Updates blog and 'About the transition from Gems to skills'): Business/Enterprise 'no earlier than March 1, 2027', when Gems can no longer be created, edited or used, 'Ask a Gem' steps in Workspace Studio stop working and remaining Gems become inactive draft skills their creators must activate; Education no earlier than June 1, 2027

#### What happened
Google moved from persona-style custom assistants (Gems) to composable skills, a pattern Claude and other agent products already used.

#### Why it matters
"Skills" is becoming a shared abstraction for packaging repeatable tasks for consumer and agent models.

#### Update, Oct 12, 2026: Spark becomes "Task" inside the main chat
9to5Google (Abner Li, Oct 12) reports a server-side rollout on Android that folds the Gemini Spark personal agent into the main Gemini app:
"Chat" and "Task" buttons sit above the prompt box, and "Task" is Spark ("New home for Spark: Start new tasks anytime and keep track of them
in your chats"). Spark leaves the navigation drawer; a new "Scheduled" page lists Spark tasks and scheduled actions (Active, Paused, Completed,
Suggested), and the conversation list mixes chats and tasks with filters (All, Ready, Needs attention, In progress). "Gemini Spark Settings"
is renamed "Task settings", which 9to5Google reads as Google de-emphasizing the Spark name. Spark still needs an AI Pro or AI Ultra plan;
iOS and the web had not changed yet. Google has not published a post about it.

Sources: [9to5Google: Gemini 'Task' redesign integrates Spark into the main chat app (Oct 12, 2026)](https://9to5google.com/2026/10/12/gemini-spark-task-redesign/); [Google Workspace Updates: Skills in the Gemini app for Workspace](https://workspaceupdates.googleblog.com/2026/09/skills-gemini-app-workspace.html); [Google Workspace Knowledge Center: About the transition from Gems to skills](https://knowledge.workspace.google.com/p/gems-migration); [Droid Life: Gemini launches skills, kills off Gems](https://www.droid-life.com/2026/09/30/gemini-launches-skills-kills-off-gems/); [TechRepublic: Google is replacing Gemini Gems with skills Nov 17](https://www.techrepublic.com/article/news-gemini-gems-skills-migration/); [Android Central: Gemini Gems are heading out](https://www.androidcentral.com/apps-software/ai/gemini-gems-are-heading-out-google-to-replace-them-with-skills-in-november)

### 2026-10-01: Anthropic launches mods for Claude Code: TypeScript plugins that change its behavior, interface and tool calls

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-10-01-claude-code-mods/

On Oct 1, 2026 Anthropic opened Claude Code to "mods": JavaScript/TypeScript event handlers shipped inside plugins that run in Claude Code's own process and can watch, rewrite or take over tool calls, prompts, commands and parts of the interface (panes, the spinner, dialogs). Users can write one or ask Claude to build it. Mods are not sandboxed and run with the user's permissions; the launch post drew ~2.1M views.

- Announced by @ClaudeDevs on Oct 1, 2026: 'You can now mod Claude Code: change how it behaves, customize the UI, swap in your own features' (~2.1M views, ~14.8k likes by Oct 2)
- A mod is a plugin whose hooks module registers handlers for events such as tool.call and ui.render; a handler can observe, rewrite or answer an event
- Can draw panes and a band above the prompt, restyle built-in UI (tool-call rows, spinner, question dialog), hold or answer a tool call, send a request to a different model, add /commands that run without a Claude turn
- Installed like any plugin (/plugin install name@marketplace); requires Claude Code v2.1.287+; on by default; works in the CLI and the Desktop app's Code tab (hooks also run in claude -p, the Agent SDK and the VS Code panel, without drawing)
- Security: mods run with the user's permissions, are not sandboxed (a process a mod starts runs outside Bash sandboxing), can read secrets and can approve tool calls an 'ask' rule would prompt for; they cannot restyle the permission prompt
- Some built-in features are now mods, e.g. /diff, AGENTS.md loading and telemetry; samples token-weather, blast-radius and replay-theater are in anthropics/claude-code-playground
- Admins can restrict mods via managed settings (e.g. allowManagedModsOnly); users can disable them with --safe-mode or disableAllHooks

#### What happened
Claude Code already had settings hooks (shell commands run on events), skills and MCP servers, all working from outside the app. Mods run
inside it, so a few lines of TypeScript can add a pane that charts context usage, stop a risky `rm -rf` behind a confirmation, or send a
request to another model. Anthropic moved some of its own features onto the same system.

#### Why it matters
Coding agents are becoming platforms that users can reprogram. Mods can approve tool calls and read secrets, which widens the supply-chain
risk of plugin marketplaces. The docs say plainly that a mod is code running with your permissions.

Sources: [@ClaudeDevs on X: You can now mod Claude Code](https://x.com/ClaudeDevs/status/2105721434807083061); [Claude Code Docs: Mods overview](https://code.claude.com/docs/en/plugins/mods/overview); [GitHub: anthropics/claude-code-playground sample mods](https://github.com/anthropics/claude-code-playground/tree/main/claude-code/mods); [GitHub: anthropics/claude-code built-in mods source](https://github.com/anthropics/claude-code/tree/main/mods)

### 2026-10-01: ChatGPT adds virtual try-on for clothing and a Favorites list for shopping

**; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-10-01-chatgpt-virtual-try-on-favorites/

On Oct 1, 2026 OpenAI launched two ChatGPT shopping features: a "Try On" button that shows uploaded selfies or full-body photos wearing clothing and accessories, built on ChatGPT Images 2.5, and Favorites, which saves products to the user's Library. It is OpenAI's second push into shopping after it dropped Instant Checkout.

- Try On: upload a selfie or full-body photo; works on shopping results or on images of items the user uploads (e.g. screenshots)
- Favorites: saved products stored in the Library alongside try-on images
- Powered by ChatGPT Images 2.5; described as a global launch (TechCrunch)

#### What happened
OpenAI added virtual try-on and saved favorites to ChatGPT's shopping results. Google has offered a similar try-on feature since 2025.

#### Why it matters
It is a commerce use of OpenAI's image model, part of OpenAI's search for consumer revenue beyond subscriptions.

Sources: [TechCrunch: ChatGPT can now virtually try on clothes for you](https://techcrunch.com/2026/10/01/chatgpt-can-now-virtually-try-on-clothes-for-you/)

### 2026-10-01: Google adds Guided Vision to Gemini Live: real-time spoken scene guidance for blind and low-vision users on Android

**; Google; Products; confidence high; https://postcutoff.com/e/2026-10-01-gemini-live-guided-vision/

On Oct 1, 2026 Google launched Guided Vision in Gemini Live, a camera mode that gives blind and low-vision users live audio descriptions and spoken cues for re-aiming the phone ("verbal reframing") while they explore their surroundings. It runs on Android 9+ wherever Gemini Live is available and was built with the visual-interpretation service Aira.

- Conversational, real-time visual assistance: dynamic audio descriptions plus spoken cues to reframe the camera
- Built for people who are blind or have low vision; Google says it also helps older adults and anyone in low light
- Android 9 and above; available wherever Gemini Live is supported; tested in India, Brazil, Singapore, Indonesia and Japan
- Access via Gemini app settings, Android Accessibility Shortcuts or TalkBack
- Partner Aira: its Trusted Tester network (1,000+ members) and specialists contributed tens of thousands of hours of visual-interpretation data and helped set safety guardrails
- Google does not name the underlying Gemini model

#### What happened
Google turned Gemini Live's camera sharing into an assistive mode. Instead of answering one-off questions about a frame, the assistant keeps
describing the scene and tells the user how to move the phone to capture what they are looking for.

#### Why it matters
Live multimodal assistants are taking over some tasks that used to need a human describer. Here a human visual-interpreter service, Aira,
supplied the training data and helped set the guardrails.

Sources: [Google blog: Guided Vision in Gemini Live, built for accessibility](https://blog.google/innovation-and-ai/products/gemini-app/guided-vision-gemini-live/)

### 2026-10-02: Meta launches Muse Gadgets: open-source ESP32 firmware and a Linux SDK for building hardware that runs its Muse agent, plus a free 'Home Link' dongle

**; Meta; Products; confidence high; https://postcutoff.com/e/2026-10-02-meta-muse-gadgets/

On Oct 2, 2026 Meta Superintelligence Labs announced Muse Gadgets, open-source ESP32 firmware and a Linux SDK that let anyone build hardware devices (e-ink displays, TV dongles, Raspberry Pi 5 projects) that work with its personal agent Muse. Meta also made 5,000 "Muse Home Link" USB-C dongles that let Muse control devices on the home network, free for Muse subscribers.

- Announced by Nat Friedman (head of product, MSL) as a 'side project' (X, ~399k views) and by Alexandr Wang (~279k views)
- Open-source ESP32 firmware plus a Linux SDK; users get an API token at gadgets.muse.ai. Suggested builds: e-ink display, HDMI TV dongle, a Muse Charm-like touchscreen; Raspberry Pi 5 supported (Engadget)
- Muse Home Link: USB-C dongle that connects Muse to smart home devices 'like TVs and speakers and anything else that exposes https'; batch of 5,000, shipping in a few weeks, free for Muse subscribers while supplies last; its code is open source too
- Wang on why open-source it: 'we want to unleash builders to explore all sorts of new and crazy ideas'
- Early builds shared on X: Federico Viticci turned an Xteink e-reader into an always-on Muse display (~52k views)
- Why open source, per Alexandr Wang: 'building gadgets is fun' and 'we want to unleash builders to explore all sorts of new and crazy ideas' (~316k views by Oct 4)
- TechCrunch (Kirsten Korosec, Oct 2): Home Link was built by the MSL team; support runs through a dedicated Discord channel. Meta presents Gadgets alongside Muse for Small Business and the new Meta Enterprise Platform unit
- Nat Friedman's Home Link post (first-hand): Home Link lets Muse talk to 'smart home devices like TVs and speakers and anything else that exposes https'; 'All the code for Muse Home Link is open source too' (~157k views by Oct 4)
- Reach of other launch posts by Oct 4: @muse official ~58k views; MSL's Jesik Min (wine-rack gadget) ~116k; MSL's David Singleton (@dps, e-ink display above his keyboard) ~81k; Wes Bos port to an RP2350 conference badge ~58k

#### What happened
Meta opened its Muse agent to hobbyist hardware. Builders flash ESP32 boards or use a Linux SDK to make devices that display or talk
to their Muse, and Meta shipped its own reference gadget, the Home Link, to connect Muse to home networks.

#### Why it matters
After its Muse Charm pocket device at Connect, Meta is pushing Muse into physical devices through an open-source maker ecosystem
rather than only its own hardware. This is a distinctive move in the race to make personal agents ambient. (Separately, MBI Deep
Dives argued on Oct 2 that Meta may never need ads to monetize Muse.)

Sources: [TechCrunch: Meta wants your next gadget to be Muse-infused](https://techcrunch.com/2026/10/02/meta-wants-you-to-build-your-own-muse-gadget/); [Nat Friedman on X: Muse Home Link (~157k views)](https://x.com/natfriedman/status/2106099384891158562); [Muse on X: Muse Gadgets are here](https://x.com/muse/status/2106119592229884289); [Muse Gadgets site](https://gadgets.muse.ai/); [Nat Friedman on X: announcing Muse Gadgets](https://x.com/natfriedman/status/2106099383037309211); [Alexandr Wang on X: Muse Gadgets](https://x.com/alexandr_wang/status/2106113742266089526); [Engadget: Meta Muse Gadgets, open-source smart home link](https://www.engadget.com/2276312/meta-muse-gadgets-open-source-smart-home-link/); [Federico Viticci on X: Xteink e-reader as a Muse display](https://x.com/viticci/status/2106139174604816765); [Alexandr Wang on X: why open-source Muse Gadgets (~316k views)](https://x.com/alexandr_wang/status/2106116686654931161)

### 2026-10-03: Google limits free Gemini app users to Flash-Lite from Oct 9; AI Plus loses Pro access

***; Google; Products; confidence high; https://postcutoff.com/e/2026-10-03-gemini-free-tier-flash-lite-only/

An updated Google support page (reported by 9to5Google on Oct 3, 2026) says that from October 9 free Gemini app users will only get the 3.5 Flash-Lite model, losing 3.6 Flash and 3.1 Pro, while $4.99/month AI Plus subscribers keep Flash-Lite and 3.6 Flash but lose Pro. AI Pro and Ultra are unchanged, and AI Pro gains Deep Think. It is a visible cut to free access to Google's stronger models. As rolled out on Oct 9, free users get an 'Auto' picker that mostly uses Flash-Lite but may route harder prompts to Flash or Pro.

- Date confirmed (9to5Google, Oct 14): Google emailed Google AI Plus ($4.99/month) subscribers that from Nov 13, 2026 Pro models can be selected in the Gemini app and AI Mode only on AI Pro and above; AI Plus keeps Flash-Lite and Flash
- Took effect Oct 9, 2026 as announced. At the same time the Gemini app added low/medium/high effort levels for each model (as in AI Studio and Antigravity). Google's help page: a higher level 'increases the model's ability to complete tasks… but uses more of' the user's compute-based limit (9to5Google; Relevant Audience). Computerworld (Oct 9): 'Gemini Pro is now only for Google AI Pro and AI Ultra subscribers'
- Effective Oct 9, 2026: free users limited to Gemini 3.5 Flash-Lite; they 'will no longer be able to access (3.6) Flash or (3.1) Pro' (Google support doc via 9to5Google)
- AI Plus ($4.99/month): Flash-Lite and 3.6 Flash only, no Pro; subscribers get an email explaining when the changes take effect
- AI Pro ($19.99/month) and AI Ultra keep full model access; AI Pro gains Deep Think
- AI Ultra costs $99.99 or $199.99/month. Until now Deep Think was limited to the two most expensive plans. The free tier had offered 3.6 Flash and varying access to 3.1 Pro (The Decoder; PPC Land per search snippet)
- The Decoder's read: real-world impact is likely small, since most casual users don't know which model they use. It may make room for the more expensive Gemini 4 Argon
- Reach: a summary post by the fan account Gemini Updates (@GeminiUpda0kkb) ('Free → Flash-Lite only • AI Plus → loses Pro models') passed 1.9M views by Oct 5 (fxtwitter)
- Second wave Oct 6: The Verge ('Google is about to remove free access to Gemini Flash and Pro'), ZDNET, Tom's Guide (suggesting free alternatives) and TechRepublic carried the change, three days before it took effect
- As rolled out (9to5Google, Oct 9): free users 'without a plan' can no longer pick a model; the picker only offers 'Auto' ('Gemini 3 models'). Google: 'Gemini will use auto model selection to send each prompt to the Gemini model that suits it best. Most prompts will go to Flash-Lite, our fastest model. Prompts that need deeper reasoning may go to Flash or Pro.' So free users can still reach Flash or Pro through routing, but not choose them. Effort levels Low / Medium / High replace the earlier 'Extended' option for subscribers
- Droid Life (Oct 15, 2026), 'Your Gemini Pro cost is going up considerably': from Nov 13, 2026 choosing Pro models in the Gemini app and in AI Mode in Search requires Google AI Pro or above; for many users that means going from about $5 to $20 a month (AI Pro's own price is unchanged)

#### What happened
Google updated its Gemini app help documentation to say that from October 9, 2026 model choice will depend on the subscription tier. Free users get only Gemini 3.5 Flash-Lite. AI Plus subscribers get Flash-Lite and 3.6 Flash. AI Pro and Ultra keep every model, and AI Pro gains Deep Think.

#### Why it matters
Free and low-tier users lose access to Google's stronger models. This is a concrete sign of inference-capacity pressure and of stronger models moving behind paywalls across consumer AI apps. The press report does not give Google's stated reason.

Sources: [Droid Life: Your Gemini Pro cost is going up considerably (Oct 15)](https://www.droid-life.com/2026/10/15/your-gemini-pro-cost-is-going-up-considerably/); [9to5Google: Gemini AI Plus subscribers will lose access to Pro models next month (Oct 14)](https://9to5google.com/2026/10/14/gemini-ai-plus-loses-pro-november/); [Relevant Audience: Gemini app limits free users to Flash-Lite from 9 October — model access by plan and effort levels](https://www.relevantaudience.com/ai/gemini-app-free-users-flash-lite-october-9/); [iThinkDiff: Gemini app drops Flash and Pro for free users on October 9](https://www.ithinkdiff.com/gemini-app-model-limits-free-ai-plus-october-9/); [The Verge: Google is about to remove free access to Gemini Flash and Pro](https://www.theverge.com/ai-artificial-intelligence/1005451/google-gemini-free-flash-lite-only); [ZDNET: Use Gemini for free? You'll soon be limited to its weakest AI model](https://www.zdnet.com/innovation/google-ai-free-model-access-limited/); [TechRepublic: Google limits free Gemini to Flash-Lite: what changes starting Oct. 9](https://www.techrepublic.com/article/news-google-gemini-free-flash-lite-access/); [Tom's Guide: Google is taking away free access to these Gemini models](https://www.tomsguide.com/ai/google-is-taking-away-free-access-to-these-gemini-models-try-these-7-free-alternatives-instead); [The Decoder: Google's new Gemini tiers cut free users to its weakest model and lock $5/month subscribers out of Pro](https://the-decoder.com/googles-new-gemini-tiers-cut-free-users-to-its-weakest-model-and-lock-5-month-subscribers-out-of-pro/); [PPC Land: Google cuts two of three Gemini models for free users from October 9](https://ppc.land/google-cuts-two-of-three-gemini-models-for-free-users-from-october-9/); [9to5Google: Gemini free users limited to Flash-Lite, Plus loses Pro starting Oct 9](https://9to5google.com/2026/10/03/gemini-model-limits-oct-26/); [9to5Google on X](https://x.com/9to5google/status/2106488912721957074); [Gemini Updates (fan account) on X (~1.9M views)](https://x.com/geminiupda0kkb/status/2106324275829281101); [9to5Google: Free Gemini users now on 'Auto' models as app adds thinking levels (Oct 9)](https://9to5google.com/2026/10/09/gemini-auto-free-users/)

### 2026-10-04: First Googlebook laptops go on sale in the US: Google's Android-based, Gemini-centred successor to the Chromebook

***; Google, Acer, ASUS, Dell, HP, Lenovo; Products; confidence high; https://postcutoff.com/e/2026-10-04-googlebook-launch/

On Oct 4, 2026 the first Googlebooks reached US buyers (Oct 5 in Canada, the UK, Ireland, France, Germany and Australia). They are laptops from Acer, ASUS, Dell, HP and Lenovo on an Android-based platform with ChromeOS elements, built around Gemini. Prices run from $899 to $1,299, with 12 months of Google AI Pro included. Gemini features include Magic Pointer (wiggle the cursor to bring Gemini to whatever is on screen), Rambler (speech to clean writing) and AI-generated widgets. Google is using the line to move its Chromebook base to Gemini.

- Timeline: unveiled May 2026 as the Chromebook's successor; pre-orders Sept 21; US shipping Oct 4, other launch countries Oct 5, 2026
- Price from $899 (Acer) to $1,299 (ASUS, HP); Dell XPS $1,199, Lenovo $1,099 (PCWorld); bundle: 12 months Google AI Pro (5TB), 3 months YouTube Premium, Adobe Photoshop, 1 year GeForce NOW
- Gemini features: Magic Pointer (wiggle the cursor to bring Gemini to on-screen content), Rambler (spoken thoughts to organised writing), Create My Widget (generated custom widgets)
- Phone integration: Continue On (resume phone tasks), phone files in the Files app, Cast My Apps (run phone apps in a desktop window)
- Hardware: Intel/Qualcomm chips with NPUs over 45 TOPS, up to 2.8K OLED touchscreens, up to 14 h battery; up to 10 years of OS updates; Titan root of trust and pKVM-isolated Linux
- Example: ASUS Googlebook 14 (Intel Core Ultra 5 325, 16GB, 512GB, 14-inch 3K 120Hz OLED) at $1,299, on sale Oct 4 at the ASUS Store and Best Buy

#### What happened
Google announced the Googlebook in May 2026 as the successor to the Chromebook. It runs on Google's Android stack with ChromeOS desktop parts and a desktop
Chrome browser. Five OEM models opened for pre-order on Sept 21 and went on sale in the US on Oct 4, 2026. Gemini is built into the interface. With Magic Pointer, the
user wiggles the cursor to ask Gemini about anything on screen. Rambler turns dictation into finished text, and Create My Widget generates custom widgets.
Every device comes with a year of Google AI Pro.

Early hands-on coverage (PCWorld, Sept 21) called the platform "made with phone-first users in mind" rather than AI-first. TechCrunch said the cursor
feature is unlikely to be the main selling point now that agents can operate computers themselves, and read the line as a way to move
Chromebook users, including the large school base, onto Gemini. Independent benchmark and battery reviews had not appeared by launch day.

#### Why it matters
Googlebook is Google's first laptop platform designed around a frontier assistant, and it bundles paid Gemini subscriptions with the hardware. It is the
PC-side counterpart to Gemini on Android, and it competes with Microsoft's Copilot+ PCs and Apple Intelligence Macs for the "AI PC".

Sources: [Google: Googlebook, the laptop your Android phone has been waiting for](https://blog.google/products-and-platforms/devices/googlebook/pre-order-googlebook/); [Googlebook product site](https://googlebook.google/); [ASUS (GlobeNewswire via Manila Times): ASUS Googlebook 14 now available in the United States (Oct 4)](https://manilatimes.net/2026/10/04/tmt-newswire/globenewswire/asus-googlebook-14-now-available-for-purchase-in-the-united-states/2438596); [TechCrunch: Google's $899 Googlebook is a bet that you'll buy a new laptop for Gemini](https://techcrunch.com/2026/09/21/googles-899-googlebook-is-a-bet-that-youll-buy-a-new-laptop-for-gemini/); [PCWorld: The first Googlebooks are here, I tried all 5](https://www.pcworld.com/article/3239947/the-first-googlebooks-are-here-i-tried-all-5-from-acer-dell-hp-asus-and-lenovo.html); [9to5Google: Google fully details Googlebooks](https://9to5google.com/2026/09/21/googlebook-launch/)

### 2026-10-04: Anthropic asks Claude users to opt in to sharing voice recordings for model training

**; Anthropic; Products; confidence medium; https://postcutoff.com/e/2026-10-04-claude-voice-data-training-opt-in/

Around Oct 4, 2026 Claude users began seeing a prompt in voice mode asking: "Allow us to use your voice data to improve our AI models." A new toggle under Settings > Privacy covers audio recordings and voice chat data from approved sessions. It is off by default, separate from the existing setting for training on chats and Claude Code sessions, and users can switch it off and delete the voice data later.

- Prompt text (BleepingComputer, Oct 4): 'Allow us to use your voice data to improve our AI models. Audio recordings and voice chat data help improve how Anthropic AI models understand and respond to speech.' Options: Allow / Not now
- Settings > Privacy: 'Allow us to use your voice data': 'For approved sessions, allow the use of your audio recordings and voice chat data to improve Anthropic AI models'
- Disabled by default and independent of the chat/coding-session training setting introduced in Anthropic's Aug 2025 consumer-terms update
- No Anthropic blog post or retention period for voice data was found as of Oct 4; BleepingComputer's report is based on the in-app prompt and a screenshot on X

#### What happened
Anthropic added a separate, opt-in consent for using voice-mode audio as training data, alongside the existing setting for text chats.

#### Why it matters
Voice recordings are biometric-adjacent data. Making the consent opt-in and separate from chat data is a stricter default than the 2025
change for chats, and it suggests Anthropic wants real speech data to improve Claude's voice models.

Sources: [BleepingComputer: Anthropic asks Claude users to share voice data for AI model training (Oct 4)](https://www.bleepingcomputer.com/news/artificial-intelligence/anthropic-asks-claude-users-to-share-voice-data-for-ai-model-training/); [Anthropic: Updates to Consumer Terms and Privacy Policy (2025, chat training opt-in)](https://www.anthropic.com/news/updates-to-our-consumer-terms); [Claude privacy center: Is my data used for model training?](https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training)

### 2026-10-04: OpenAI's Thibault Sottiaux pledges a clear Codex/ChatGPT Work improvement every day for 28 days, or a 'full reset' (~2.2M views)

**; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-10-04-codex-28-day-daily-improvement-pledge/

On Oct 4, 2026 Thibault Sottiaux, who leads OpenAI's core product and platform, posted that for the next 28 days OpenAI will ship each day either "one thing that is a clear improvement and relevant for most codex/work users" or "a full reset". The post reached ~2.2M views. Crypto and Chinese tech outlets read it as a response to Claude Code's momentum after Claude Opus 5.5.

- Post (Oct 4, 20:33 UTC): 'Over the next 28 days, each day we'll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset. Let the improvements begin.' (~2.17M views, ~19k likes by Oct 5, fxtwitter)
- Window: Oct 4 – Nov 1, 2026 (PANews, KuCoin)
- Reported focus areas: simplifying the product, efficiency (more usable capacity per user), breakthrough features and new models (PANews, citing Sottiaux)
- 'Full reset' most likely means resetting users' usage limits, which OpenAI has done for Codex and ChatGPT Work before. That reading is our inference; the post does not define it

#### What happened
Sottiaux, formerly the Codex lead, announced a 28-day run of daily shipping for Codex and ChatGPT Work on X. On any day without a
user-visible improvement, users get a "full reset" instead.

#### Why it matters
It is a public commitment to a fast shipping pace in the coding-agent race, a week after DevDay and amid Claude Code vs Codex
comparisons following Opus 5.5. Later runs can check whether the daily cadence held through Nov 1.

Sources: [Thibault Sottiaux on X (Oct 4)](https://x.com/thsottiaux/status/2106845241357824205); [PANews: Codex lead Tibo: daily iterations for 28 days, or a full reset](https://panews.io/articles/01a10a03-e793-7231-9b25-031d1c7bf409); [KuCoin News: OpenAI sets 28-day sprint for Codex with daily improvements or resets](https://www.kucoin.com/news/flash/openai-sets-28-day-sprint-for-codex-with-daily-improvements-or-resets)

### 2026-10-05: Anthropic launches in-country Claude inference in India via Amazon Bedrock

**; Anthropic, Amazon; Products; confidence medium; https://postcutoff.com/e/2026-10-05-claude-india-in-country-inference-bedrock/

On Oct 5, 2026 Anthropic said Claude requests sent to an India endpoint on Amazon Bedrock are now processed on servers in India. This meets data-residency needs of Indian banks, insurers, government bodies and large enterprises. It delivers on an August 2026 promise by Anthropic India MD Irina Ghose. Reliance and CRED tested it in private preview.

- Models listed in Indian press: Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5. The 5.5 models are not mentioned; not checked against AWS docs
- Irina Ghose (MD, Anthropic India): 'In August, we committed to bringing Claude inference to India, and we are delivering on that promise' and 'Our customers, especially in regulated sectors and government, want to use Claude on their most critical work, with requests processed in the country.'
- Private-preview customers: Reliance, CRED and others; Indian press lists Claude users including Kotak, Axis, IndusInd, NPCI, TCS (50,000 associates), Infosys, Cognizant, Swiggy, Godrej and Freshworks
- Partner push: first Claude Certification for Partners held in Bengaluru (Aug 2026), aiming to certify up to ~5,000 people (afaqs, CRN Asia)

#### What happened
Anthropic and AWS made India-resident Claude inference generally available on Bedrock. We did not find an anthropic.com or AWS post for the
launch, so the details come from Indian trade press that appears to work from an Anthropic release.

#### Why it matters
India is one of Claude's largest markets, and data-residency rules have kept regulated sectors from production use. Local inference is part of the
sovereign-AI push that competing labs (OpenAI, Google) are also making in India.

Sources: [afaqs!: Anthropic brings Claude to India with in-country inference via Amazon Bedrock (Oct 5)](https://www.afaqs.com/news/digital/anthropic-brings-claude-to-india-with-in-country-inference-via-amazon-bedrock-12623986); [Storyboard18: Anthropic brings Claude in-country inference to India through Amazon Bedrock](https://www.storyboard18.com/how-it-works/anthropic-brings-claude-in-country-inference-to-india-through-amazon-bedrock-111980.htm); [CRN Asia: Anthropic expands India strategy with local Claude inference and partner ecosystem investments](https://www.crnasia.com/india/news/2026/anthropic-expands-india-strategy-with-local-claude-inference-and-partner-ecosystem-investments); [Anthropic: Anthropic appoints Irina Ghose as Managing Director of India](https://anthropic.com/news/anthropic-appoints-irina-ghose-as-managing-director-of-india)

### 2026-10-06: OpenAI opens the Decisions API to all developers in public beta: gpt-6-luna typed answers at $0.10 per 1M input tokens, output free

***; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-10-06-openai-decisions-api-public-beta/

On Oct 6, 2026, a week after announcing it at DevDay, OpenAI released the Decisions API (POST /v1/decisions) in public beta. It runs gpt-6-luna and returns typed answers (a probability, a choice from fixed options, or a rubric score) about 10x faster than the Responses API. It costs $0.10 per 1M input tokens with no output or caching charges, and OpenAI expects general availability "in the coming weeks". It is OpenAI's answer to TypeSafe's Jev and the decision models that Cloudflare and Perplexity shipped in the meantime.

- API changelog (Oct 6): 'Released the Decisions API in beta with gpt-6-luna. Turn text and images into typed answers 10x faster than the Responses API.'
- Endpoint: POST https://api.openai.com/v1/decisions; gpt-6-luna is the only supported model; Playground at platform.openai.com/decisions
- Three question types: predicate (probability 0–1 that a condition is true), choice (one of the supplied values, with a probability distribution and a confidence field) and score (probability-weighted average over ordered rubric levels); several independent questions can share one input
- Input: text, or user messages with text and images
- Price: $0.10 per 1M input tokens; no output-token, cache-read or cache-write charges; regional-processing premiums and long-context multipliers apply
- Supports Zero Data Retention and HIPAA for eligible customers; data residency in the US and Europe (EEA + Switzerland)
- Status: public beta, GA expected 'in the coming weeks'; SDK support from Python 3.26.0, JavaScript 7.30.0, Go 3.73.0, Ruby 0.101.0, Java 4.78.0
- Comparison (Simon Willison): Jev charges 4.2 cents per 1M input tokens, OpenAI 10 cents; Perplexity's Decisions API is $0.04; unlike Jev, gpt-6-luna takes images
- Same day, OpenAI also cut API usage tiers from five to three (Build, Launch, Grow); on Oct 5 it added an in-product HIPAA BAA flow for API organizations
- General availability (Oct 22, 2026): @OpenAIDevs announced the Decisions API is 'now generally available' for classifying text and images, evaluating inputs and choosing what happens next 'without generating a full model response' (~127k views); TechCrunch framed it as OpenAI's Jev competitor going wide as decision-model competition heats up

#### What happened
OpenAI announced the Decisions API at DevDay on Sept 29, 2026, in limited preview, and promised a broad release "in the coming days"
without giving a price. On Oct 6 the API changelog listed the Decisions API as released in public beta. The new guide explains how it works.
A request carries shared `input` (text and/or images) and an array of named `questions`, each of type `predicate`, `choice` or `score`.
The response returns an `answers` array with probabilities rather than generated text, plus a refusal type. OpenAI says this is about
10x faster than the Responses API. OpenAI recommends setting thresholds with labelled examples from your own application. Only input
tokens are billed, at $0.10 per 1M, which is gpt-6-luna's normal input rate.

Simon Willison released an `llm-openai-decisions` plugin the same day. He noted that the API shape is "very similar to Jev" (TypeSafe's decision model)
and that OpenAI's input price is about 2.4x Jev's.

#### Why it matters
Decision models (one forward pass that scores a fixed set of answers) went from one startup's product (Jev, Sept 15) to a category with
OpenAI, Cloudflare and Perplexity offerings within three weeks. OpenAI's version is the most expensive per input token of the four, but it
accepts images and supports ZDR, HIPAA and EU data residency, which matter to the enterprise moderation, routing and triage work these
models target. No accuracy benchmarks or maximum number of options were published.

Sources: [OpenAI Devs on X: Decisions API generally available (Oct 22)](https://x.com/OpenAIDevs/status/2113346642787086538); [TechCrunch: OpenAI makes its Jev competitor widely available as decision model competition heats up](https://techcrunch.com/2026/10/22/openai-makes-its-jev-competitor-widely-available-as-decision-model-competition-heats-up/); [OpenAI API changelog (Oct 6: Decisions API beta; usage tiers)](https://developers.openai.com/api/docs/changelog); [OpenAI docs: Decisions guide (question types, pricing and availability)](https://developers.openai.com/api/docs/guides/decisions); [OpenAI: DevDay 2026 recap (Decisions API announced Sept 29)](https://openai.com/index/devday-2026-recap/); [Simon Willison: llm-openai-decisions 0.1a0 (Oct 6)](https://simonwillison.net/2026/Oct/6/llm-openai-decisions/)

### 2026-10-06: Anthropic brings Claude into Google Docs, Sheets and Slides: a Workspace add-on sidebar plus connectors that edit Google files from Claude (beta)

**; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-10-06-claude-for-google-workspace/

On Oct 6, 2026, Anthropic launched Claude for Google Workspace in beta for Pro, Max, Team and Enterprise plans. It is a Marketplace add-on that opens Claude in a sidebar inside Google Docs, Sheets and Slides and edits the open file in place. Anthropic also launched Google Docs, Sheets and Slides connectors that let Claude create and edit Google files from a chat. It is the Google counterpart of Claude for Microsoft 365.

- Add-on: one Google Workspace Marketplace install covers Docs, Sheets and Slides (Extensions > Claude > Open Claude); admins can deploy it to a whole domain or to selected groups
- Claude sees the open file and the selected text, cells or slides. It edits in place (changes show in version history) or proposes bigger rewrites as suggestion cards
- Sheets: formulas, pivot tables, native charts and new tabs; it can move a range into Python for joins and cleaning and write the results back. Slides: builds slides from the deck's own layouts and themes and flags overlapping or unreadable elements
- Two modes: 'Ask before edits' (default, an approval card per change) and 'Accept all edits'
- The sidebar uses the same models, connectors and skills as the user's Claude account. On Enterprise, the Compliance API, CMEK and OpenTelemetry audit export also apply
- Separate Google Docs, Sheets and Slides connectors (beta) let Claude create or edit Google files from claude.ai, opening them in a pane beside the chat; on Team and Enterprise an owner must enable them first
- Plans: Pro, Max, Team and Enterprise (beta)

#### What happened
Anthropic published "Claude now works in Google Docs, Sheets, and Slides" on Oct 6, 2026. The launch has two parts, both in beta:

1. **Claude for Google Workspace add-on.** Claude opens in a sidebar next to the file. It reads the document, spreadsheet or deck, sees the
   current selection and makes changes in place. Larger rewrites appear as suggestion cards that highlight the passage they would change.
   By default each edit needs approval ("Ask before edits"); an "Accept all edits" mode lets it work through a task without stopping.
2. **Google Docs, Sheets and Slides connectors.** These work the other way: from a chat in Claude, the user pastes a Google file link or asks for a new doc,
   sheet or deck. Claude creates or edits it, and on supported setups the file opens in a pane beside the conversation.

The help center says the add-on is different from the older Google Workspace connectors. Those only let Claude search Drive, Gmail and Calendar.

#### Why it matters
Anthropic already sells Claude inside Microsoft 365 (Excel, PowerPoint, Word, Outlook). This launch brings the same in-file agent to
Google Workspace, Microsoft's main office-suite rival and the home of Google's own Gemini side panel. Claude now competes with Gemini directly inside Google's
apps. Enterprise controls (Compliance API, CMEK, audit export) come with it, which matters for company-wide deployment.

Sources: [Claude blog: Claude now works in Google Docs, Sheets, and Slides (Oct 6)](https://claude.com/resources/articles/claude-now-works-in-google-docs-sheets-and-slides); [Claude Help Center: Use Claude in Google Docs, Sheets, and Slides](https://support.claude.com/en/articles/16951679-use-claude-in-google-docs-sheets-and-slides); [Google Workspace Marketplace: Claude](https://workspace.google.com/marketplace/app/claude/12459801340)

### 2026-10-06: Google Earth AI for public health: WHO AFRO's Ebola team used a Geospatial Reasoning agent to find 48 exposed settlements in the DRC outbreak

**; Google, Google Research, WHO; Products; confidence high; https://postcutoff.com/e/2026-10-06-google-earth-ai-who-ebola-geospatial-agent/

Google said on Oct 6, 2026 that WHO's Africa regional office used its Geospatial Reasoning agent prototype (Google Earth AI) during the DRC Ebola outbreak to pinpoint 48 exposed settlements and more than 45,500 at-risk people, work that normally takes weeks. Google also made Population Dynamics Insights (PDFM) commercially available in preview on Google Maps Platform.

- Post: 'Making global public health more proactive with Google Earth AI', Oct 6, 2026, by Shravya Shetty (Google Research) and Monica Bharel (Google Health)
- WHO AFRO's Emergency Preparedness and Response Hub: 'pinpointed 48 exposed settlements and located more than 45,500 at-risk people — a process that normally would have taken weeks'; used to place mobile labs and plan border surveillance
- Population Dynamics Insights (PDFM) now commercially available via Google Maps Platform in Preview, free for eligible researchers; Google Earth credits for noncommercial public-health work
- Google.org funding for the DRC's National Institute of Biomedical Research (INRB) to modernize disease surveillance

#### What happened
The same Bundibugyo Ebola outbreak in which WHO AFRO uses Claude for situation reports also saw Google's geospatial agent used to map
exposed communities from satellite and mobility data.

#### Why it matters
Two frontier labs' tools are now documented in one live outbreak response, a sign that AI agents are entering routine emergency
public-health work. The numbers are Google's own; no independent evaluation was published.

Sources: [Google blog: Making global public health more proactive with Google Earth AI (Oct 6, 2026)](https://blog.google/innovation-and-ai/technology/health/google-earth-ai/)

### 2026-10-07: OpenAI brings GPT-6 to all ChatGPT users with 'Intelligent UI' answers built from interactive components

****; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-10-07-gpt-6-chatgpt-intelligent-ui/

On Oct 7, 2026 OpenAI replaced GPT-5.6 in ChatGPT's Chat tab with October updates of GPT-6 Sol (Plus, Pro, Business, Enterprise) and GPT-6 Luna (Free and Go, from Oct 8), bringing GPT-6 to its more than 1.2 billion weekly users. The models introduce "Intelligent UI": answers composed from text plus graphics, charts, maps, forms, tappable buttons and small interactive tools, rendered progressively from a library of native components. ChatGPT can also start answering while it keeps thinking. Work and Codex keep the September model versions.

- Rollout: Oct 7, 2026 to Plus, Pro, Business and Enterprise in the Chat tab; Free and Go from Oct 8; Enterprise subject to admin settings (OpenAI)
- Models: GPT-6 Sol for paid tiers and GPT-6 Luna for Free/Go, 'both tuned for everyday conversation'; they replace GPT-5.6 Sol and GPT-5.6 Luna in ChatGPT. Work and Codex keep the September versions (OpenAI on X, ~320k views; system card)
- Intelligent UI: GPT-6 was trained to compose answers from text, visuals and interactive elements; OpenAI built 'a library of native, streamable components, along with a compiler that processes the interface as the model generates it', so the interface appears progressively. Examples: Sunday-roast timing next to a recipe, road-trip stops on a map, a savings calculator, a bill splitter, a game in the chat
- Interleaved thinking and answering: ChatGPT can begin answering while it continues to think. OpenAI: on an internal eval of everyday agentic tasks, GPT-6 Extra High starts answering as fast as GPT-5.6 Medium while scoring better than GPT-5.6 Extra High; for questions needing web search, GPT-6 Instant starts answering 44% sooner on average than GPT-5.6 Instant
- System card 'GPT-6 Sol and GPT-6 Luna: October 2026 update' (Deployment Safety Hub, Oct 7): treated as High capability in Cybersecurity and in Biological and Chemical under the Preparedness Framework, below High in AI Self-Improvement; same safeguards as GPT-5.6 Sol/Luna; stronger resistance to multi-turn jailbreaks and less dishonesty and guardrail circumvention than GPT-5.6, with regressions judged 'generally low severity' after manual review
- API: the chat-latest snapshot was updated on Oct 7 to point to the latest ChatGPT model for paid users; OpenAI recommends the GPT-6 family for production API use (API changelog)
- TechCrunch quotes ChatGPT product manager Aarush Selvan: 'the most helpful answers aren't just text'; users can adjust how many visuals appear. Hacker News thread reached ~740 points

#### What happened
A month after GPT-6 Sol and Luna launched for paid users in ChatGPT Work, Codex and the API, OpenAI moved the everyday Chat tab to
GPT-6. The versions shipped in Chat are new October checkpoints, tuned for conversation; the system card names them "October" to
tell them apart from the September versions that still run in Work and Codex. The headline feature is Intelligent UI: instead of
always replying in prose, the model decides when a comparison table, interactive diagram, map, form or small tool would work better,
and the interface streams in as it is generated. OpenAI also trained the model to interleave thinking and answering, writing partial
answers while it keeps reasoning, to cut waiting time.

#### Why it matters
This is the first time GPT-6-class models reach ChatGPT's free users, and it changes what a chatbot answer looks like for its 1.2
billion weekly users. OpenAI frames it as a step from fixed software toward interfaces generated per request ("Instead of people
adapting to software, software will adapt to people"). It also makes the free-tier model a Preparedness "High" model in cyber and
bio, with the safeguards that implies.

Note: benchmark claims are OpenAI's internal evaluations; the Verge article was not read (summary from Techmeme/sweep only).

Sources: [OpenAI: GPT-6 and Intelligent UI for everyone](https://openai.com/index/gpt-6-for-everyone/); [OpenAI Deployment Safety Hub: GPT-6 Sol and GPT-6 Luna, October 2026 update (system card)](https://deploymentsafety.openai.com/gpt-6-october); [OpenAI on X: GPT-6 with Intelligent UI rolls out globally](https://x.com/OpenAI/status/2107895006350791071); [OpenAI API changelog (chat-latest updated Oct 7)](https://developers.openai.com/api/docs/changelog); [TechCrunch: ChatGPT is getting a lot more visual, with the launch of a new interface](https://techcrunch.com/2026/10/07/chatgpt-is-getting-a-lot-more-visual-with-the-launch-of-a-new-interface/); [The Verge: ChatGPT's 'Intelligent UI' update fills its responses with pictures, charts, and buttons](https://www.theverge.com/ai-artificial-intelligence/1007276/openai-chatgpt-intelligent-ui-gpt-6); [Unite.AI: OpenAI brings GPT-6 with Intelligent UI to all ChatGPT tiers](https://www.unite.ai/openai-brings-gpt-6-with-intelligent-ui-to-all-chatgpt-tiers/); [Hacker News discussion](https://news.ycombinator.com/item?id=49996425)

### 2026-10-07: Google launches Playground, an experimental no-code platform for making and sharing games from text prompts

**; Google; Products; confidence high; https://postcutoff.com/e/2026-10-07-google-playground-ai-game-maker/

On Oct 7, 2026 Google launched Playground (playground.google), an experimental web platform where US users aged 18+ create browser games by describing them in prompts, then play, rate and share them. Per The Verge/Techmeme it is powered by Gemini, Nano Banana (images) and Lyria (music); a Unity Spark integration for more advanced building is in testing.

- Launched Oct 7, 2026 for US users aged 18+ at playground.google; creation access is tiered by Google AI subscription (Google blog)
- Text-prompt game creation without code; games play in the browser on any device; community gallery with ratings, multiplayer and in-game leaderboards; private, link or public sharing; tied to Play Games profiles
- Models: Gemini, Nano Banana and Lyria (Techmeme/The Verge summary; the Google post does not name models)
- Upcoming integration with Unity Spark for professional-level building features, in testing
- Published games pass safety screening against community guidelines, and users can report them

#### What happened
Google added a consumer game-making product to its AI line-up. Users describe a game in a few prompts, and Playground builds a
playable browser game that can be published to a community gallery. The blog post is by Maryam Karimzadehgan (AI Innovation + Research).

#### Why it matters
It is another step in "vibe-coding" for consumers. Google offers it as a full stack with its own text, image and music models, plus distribution.

Sources: [Google blog: Introducing Playground: Create and play custom games](https://blog.google/innovation-and-ai/technology/ai/playground-experimental-gaming-platform/); [The Verge: Google launches Playground (Unity Spark, AI)](https://www.theverge.com/tech/1006477/google-playground-unity-spark-ai)

### 2026-10-08: Anthropic launches Claude Dashboards (live BI dashboards from plain-language questions) and Claude Motion (code-animated MP4 explainers); Docs, Slides and Design leave beta for all plans

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-10-08-claude-dashboards-motion-docs-ga/

On October 8, 2026 Anthropic added two beta products to Claude: Dashboards, which connects to data warehouses and CRMs (BigQuery, Databricks, Snowflake, Salesforce, Redshift, ClickHouse) and builds live dashboards from plain-language questions, and Motion, which turns reports, charts and images into editable animations exported as MP4 by writing animation code rather than using a video-generation model. Claude Docs, Slides and Design came out of beta on every plan, including Free, after users made more than 45 million documents, decks and designs.

- Claude Dashboards: beta on paid plans; connectors to Amazon Redshift, BigQuery, ClickHouse, Databricks, Snowflake and Salesforce; dashboards stay current as data changes; export to Amplitude, Grafana, Hex, Mixpanel, PostHog, Sigma and others (Anthropic, RuntimeWire)
- Claude Motion: beta on Team and Enterprise (an admin must enable it); e.g. 'turn a quarterly report into a 30-second explainer'; exports MP4; editing hand-offs to Adobe, Descript, HeyGen, Runway and others; animates with code rather than a video-generation model (Anthropic, RuntimeWire)
- Docs, Slides and Design are generally available on all plans including Free; more than 45 million docs, decks and designs had been made in Claude (Anthropic)
- The standalone Claude Design site (claude.ai/design) closes on Dec 14, 2026; Design now works inside regular Claude conversations
- Also: customer-managed encryption keys (CMEK) for Artifacts, and team and external sharing of documents (Anthropic)
- Reuters framed it as AI labs racing to broaden workplace productivity suites

#### What happened
Anthropic announced Claude Dashboards and Claude Motion on October 8, 2026
([Claude](https://claude.com/resources/articles/dashboards-and-motion);
[Reuters via MarketScreener](https://www.marketscreener.com/news/anthropic-launches-dashboard-animation-tools-for-claude-ce785ddfdd8bf023)).
Dashboards lets a user ask a business question in plain language; Claude queries the connected warehouse or CRM, builds a dashboard
and keeps it current. Motion takes text, charts, shapes and images and produces editable animations, exported as MP4 files, by
generating animation code. Anthropic stresses that this is not a video-generation model.

The same day, the Docs, Slides and Design tools launched in beta in September became generally available on every plan, including
Free, and the separate Claude Design site is to close on December 14, 2026.

#### Why it matters
It pushes Claude further into business-intelligence and presentation tools (Tableau, Looker, PowerPoint, Canva), the same office
suite race OpenAI, Google and Meta are running. Motion's code-based approach means the result is editable and deterministic, unlike
generative video.

Sources: [Claude: Dashboards and Motion](https://claude.com/resources/articles/dashboards-and-motion); [Reuters (via MarketScreener): Anthropic launches dashboard, animation tools for Claude](https://www.marketscreener.com/news/anthropic-launches-dashboard-animation-tools-for-claude-ce785ddfdd8bf023); [RuntimeWire: Anthropic adds live dashboards and editable animations to Claude](https://runtimewire.com/article/anthropic-claude-dashboards-motion-beta); [Reuters: Anthropic launches dashboard, animation tools for Claude](https://www.reuters.com/technology/anthropic-launches-dashboard-animation-tools-claude-2026-10-08/)

### 2026-10-08: OpenAI ships Ultrafast mode for GPT-6.1 Sol in the API, Codex and ChatGPT Work

***; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-10-08-gpt-6-1-sol-ultrafast/

On Oct 8, 2026, nine days after promising it "in the coming days", OpenAI made its Ultrafast service tier available for GPT-6.1 Sol to all API users (service_tier "ultrafast" in the Responses API) and in Codex and ChatGPT Work for Pro $500 and eligible Enterprise and Edu plans. Ultrafast costs six times the standard price: $12 input and $60 output per 1M short-context tokens.

- API changelog (Oct 8): 'Added Ultrafast mode for GPT-6.1 Sol in the Responses API. Use gpt-6.1-sol with service_tier: "ultrafast" to reduce the time between generated output tokens.' Available to all API users subject to rate limits, with global processing and US and EU data residency
- Pricing (per 1M tokens, short context): $12 input, $0.60 cached input, $15 cache writes, $60 output, 6x the standard $2/$0.10/$2.50/$10; long context $24/$1.20/$30/$90. GPT-6 Astra Ultrafast is $60/$300 (OpenAI pricing page)
- Ultrafast has separate rate limits from Standard and Fast; OpenAI strongly recommends WebSockets for agentic use (Ultrafast guide)
- Codex and ChatGPT Work (Codex changelog, Oct 8): available on Pro $500 and eligible Enterprise and Edu plans; Enterprise access is off by default and must be enabled by workspace owners; US and Europe inference residency
- At GPT-6.1 Sol's Sept 29 launch OpenAI promised up to 8x faster token generation in Codex and up to 6x via the API (OpenAI; third-party summaries cite up to ~300 tokens/s)

#### What happened
OpenAI introduced Ultrafast as a premium inference tier with GPT-6 Astra and promised a GPT-6.1 Sol version at DevDay (Sept 29).
It shipped on Oct 8 in both the API and Codex/ChatGPT Work. In the API anyone can use it; in Codex it is limited to the $500 Pro
plan and enterprise and education workspaces that enable it.

#### Why it matters
Speed is becoming a priced product dimension: the same model now comes in Standard, Fast (2x) and Ultrafast (6x) tiers. For agentic
coding, where runs take many sequential tool calls, faster token generation shortens wall-clock time, which OpenAI sells at a
premium on its top plan.

Sources: [OpenAI API changelog (Oct 8: Ultrafast for GPT-6.1 Sol)](https://developers.openai.com/api/docs/changelog); [OpenAI docs: Ultrafast mode](https://developers.openai.com/api/docs/guides/ultrafast-mode); [OpenAI API pricing (Ultrafast tab)](https://developers.openai.com/api/docs/pricing?latest-pricing=ultrafast); [Codex changelog (Oct 8: GPT-6.1 Sol Ultrafast in Codex and ChatGPT Work)](https://developers.openai.com/codex/changelog); [OpenAI: Introducing GPT-6.1 Sol (Sept 29)](https://openai.com/index/introducing-gpt-6-1-sol/)

### 2026-10-08: Tencent to shut its Yuanqi agent-building platform on Nov 9, two weeks after announcing the end of its QClaw assistant

**; Tencent; Products; confidence medium; https://postcutoff.com/e/2026-10-08-tencent-shuts-yuanqi-qclaw/

In a notice dated Oct 8, 2026 Tencent said Yuanqi, its no-code platform for building and distributing AI agents (launched May 2024), will stop service on Nov 9, 2026, citing "business adjustments". Agents published from Yuanqi to WeChat accounts will stop replying. On Sept 24 Tencent had said its QClaw AI assistant would close on Dec 24. Tencent is steering users to its WorkBuddy office agent.

- Yuanqi: from Oct 29 no new agents or API keys; on Nov 9, 2026 app access, data reads and writes, API calls and the developer console are switched off (Pandaily, citing Tencent's Oct 8 notice)
- Agents published from Yuanqi to WeChat Official Accounts, Service Accounts and WeChat Customer Service stop answering on Nov 9
- Data export: the export tool works until Nov 9; an export-only console stays open until Jan 9, 2027, after which user data is deleted (except legally required logs)
- Yuanqi was unveiled on May 17, 2024 at a Tencent Cloud generative-AI summit as a one-stop platform to create agents without code and distribute them to QQ, WeChat and Tencent Cloud
- QClaw (TechNode, Sept 24, citing Electronic Engineering Journal): registrations and subscriptions stopped Sept 24; service ends at midnight on Dec 24, 2026; backups downloadable until Mar 24, 2027; Tencent offers a migration path to WorkBuddy with points as an incentive
- Tencent gave no reason beyond 'business adjustments' for either closure and named no successor for Yuanqi's distribution role

#### What happened
Tencent is closing two of its agent products within three months. Yuanqi let small businesses and creators build a chatbot with prompts,
plugins and workflows and place it inside WeChat accounts. Operators that rely on such bots have about four weeks to rebuild them
elsewhere. QClaw, an AI assistant, stopped taking new users on Sept 24 and closes on Dec 24.

Pandaily notes that Tencent's agent work has moved toward products it runs itself, such as the WorkBuddy office agent and vertical
agents inside its own apps. We did not find Tencent's original notices; both closures are reported by trade press.

#### Why it matters
It shows a large platform retiring the 2024-era "build your own agent" store model in favour of first-party agents, and it removes a
common way to run customer-service bots on WeChat.

Sources: [Pandaily: Tencent will shut down its Yuanqi AI agent platform on November 9](https://pandaily.com/tencent-yuanqi-ai-agent-platform-shutdown-november-9-data-export-january-2027); [TechNode: Tencent's QClaw to stop operating on December 24](https://technode.com/2026/09/24/tencents-qclaw-to-stop-operating-on-december-24/)

### 2026-10-09: ByteDance merges its office agent TraeWork and its coding tool TraeCode into one TRAE app with Agent and IDE modes

**; ByteDance; Products; confidence medium; https://postcutoff.com/e/2026-10-09-bytedance-merges-traework-traecode-trae/

On Oct 9, 2026 ByteDance announced that TraeWork, its AI agent for office work, and TraeCode, its AI coding tool, are now one product called TRAE. The app has an Agent mode for managing tasks across projects and an IDE mode for detailed coding and debugging, and runs on desktop, web and mobile. The standalone TraeWork service shuts down on Oct 23, 2026.

- Announced Oct 9, 2026 (TechNode, Oct 10, citing TRAE's Chinese documentation on data migration)
- One app for task planning, content creation and software development; two modes: Agent mode for 'managing tasks across projects' and IDE mode for coding and debugging
- Desktop, web and mobile: a task started on a computer can be followed or given new instructions from a phone
- Migration: TraeWork users move chat history, account information and local data through an in-app upgrade; 'The standalone TraeWork service will shut down on October 23'
- Not reported: which models power it, pricing and user numbers
- Context (search summary, not verified): August reports said ByteDance planned to fold the TRAE and Coze teams into the Doubao organization; how the TRAE brand relates to a planned Doubao Work app is not explained

#### What happened
ByteDance had two separate agent products under the Trae name: TraeCode, an AI coding environment, and TraeWork, a general agent for
office tasks. It has now combined them into one app, in which the same agent plans tasks, writes documents and writes code, with a
full IDE view available when the user wants to work in the code directly.

Only TechNode's short report was read; ByteDance's own announcement was not found.

#### Why it matters
ByteDance is treating coding and general office work as one agent product instead of two. The change comes in the same week as
personal-agent launches by Ant Group and Alibaba.

Sources: [TechNode: ByteDance merges TraeWork and TraeCode into unified TRAE platform (Oct 10, 2026)](https://technode.com/2026/10/10/bytedance-merges-traework-and-traecode-into-unified-trae-platform/)

### 2026-10-09: Google stops selling Gemini Code Assist Standard and Enterprise and points customers to Antigravity

**; Google; Products; confidence high; https://postcutoff.com/e/2026-10-09-google-ends-gemini-code-assist-sales-antigravity/

From Oct 9, 2026 new Gemini Code Assist Standard and Enterprise subscriptions can no longer be purchased. Existing subscriptions auto-renew through the rest of 2026, but auto-renewal ends in 2027. Google's release note sends developers to Antigravity, its agentic coding environment, available through eligible Gemini Enterprise subscriptions and the Gemini Enterprise Agent Platform.

- Official release note (Oct 9, 2026): 'Sale of Gemini Code Assist subscriptions is ending … Existing subscriptions will continue to auto-renew through the rest of 2026, but auto-renewal will end in 2027'
- Earlier step (Sept 4, 2026): new subscriptions could no longer be bought in the Cloud console by billing accounts without one; only via Google Cloud sales
- Replacement: Antigravity through eligible Gemini Enterprise subscriptions and Gemini Enterprise Agent Platform; the note gives no transition pricing or migration tooling (AI Weekly)
- Secondary sources give Jan 31, 2027 as the last date to add licences and renew annual plans; we could not confirm this in the official note
- Reported widely only on Oct 16 (TechInformed: 'Google ends Gemini Code Assist sales, sets renewal deadlines')

#### What happened
Google is folding its IDE-assistant product into its agent-first coding tool. Gemini Code Assist, launched as Google's answer to
GitHub Copilot, stops being sold to new enterprise customers; Antigravity becomes the recommended tool.

#### Why it matters
It is a sign of the industry-wide move from autocomplete-style assistants to coding agents: a major vendor retiring its Copilot-style
product rather than upgrading it.

Sources: [Google Cloud: Gemini Code Assist release notes (Oct 9, 2026)](https://docs.cloud.google.com/gemini/docs/codeassist/release-notes); [AI Weekly: Google ends Gemini Code Assist sales, points buyers to Antigravity](https://aiweekly.co/alerts/google-ends-gemini-code-assist-sales-points-buyers-to-antigravity)

### 2026-10-13: Apple announces HomeView, its first smart display, plus a new Apple TV 4K and HomePod mini 2, all built around Siri AI; HomeView will not be sold in the EU

***; Apple; Products; confidence high; https://postcutoff.com/e/2026-10-13-apple-homeview-siri-ai-home-devices/

On Oct 13, 2026 Apple announced three home devices by press release: Apple HomeView, a $299 7-inch smart display (tabletop and wall-mounted, A18 chip, ships Nov 13); the fourth-generation Apple TV 4K ($299, A19, three far-field microphones, ships Nov 13); and HomePod mini 2 ($149, S10, ships Oct 30). All are sold around Siri AI, the rebuilt assistant that shipped in beta with iOS 27. Siri AI is in beta in HomeView software and tvOS 27.2, comes to HomePod mini 2 in a later update, is English-only in eight countries, and needs a paired Apple Intelligence iPhone or iPad for personal requests. Apple will not sell HomeView in the EU because Siri AI is unavailable in iOS there.

- HomeView: $299 for either model (tabletop in four colors; wall-mounted in two); 7-inch Liquid Retina display; A18 chip with Secure Exclaves; pre-orders Oct 13 in Australia, Canada, India, New Zealand, the UK and the US (Singapore from Oct 15; South Africa via resellers); available Fri Nov 13, 2026
- HomeView users: recognizes up to six people, each with a personal Home Screen. Personal content (Messages, Calendar) appears only when the enrolled user is recognized by face and their paired iPhone or iPad is present and recently unlocked. John Gruber: 'HomeView does not have FaceID'; an Apple Watch does not count as the personal device
- Siri AI on HomeView (Apple): 'broad world knowledge', onscreen awareness and music discovery for anyone; personal context and actions across apps only with a paired Apple Intelligence-enabled iPhone or iPad. Footnote: Siri AI 'is in beta' in HomeView software 27.2; 'some features that rely on server-side models have daily usage limits', and expanded access will cost a fee in the future
- Models: Apple names none. It says Siri requests run on-device or in Private Cloud Compute. Gemini is not mentioned in the Oct 13 releases (background: Apple said in 2026 that its Siri AI models were built with Google's Gemini; see the WWDC 2026 and iOS 27 entries)
- Languages and countries for Siri AI on these devices: English in the US, Australia, Canada, India, New Zealand, Singapore, South Africa and the UK; Canadian French later
- EU: HomeView is 'not available for sale' in the European Union because Siri AI is currently unavailable in iOS there; Apple says it is 'working constructively with the EU in an attempt to find a path forward' and gives no date (Apple press release; MacRumors)
- Apple TV 4K (4th generation): $299; A19 chip (Apple: up to 50% faster CPU, up to 2x GPU), dedicated AV1 decoder, three far-field microphones for hands-free 'Siri'; 128GB storage, Wi-Fi 7 and Bluetooth 6 per The Verge (via Techmeme's summary; article not read); pre-orders Oct 13 in the US and 29 other countries, in stores Nov 13
- Siri AI on Apple TV: full-screen conversational search and recommendations, questions during playback, smart-home control; beta in tvOS 27.2 on Apple TV 4K (2nd generation or later); the home owner needs an Apple Intelligence-enabled iPhone or iPad with Siri AI turned on
- HomePod mini 2: $149, S10 chip, 'Sound Glow' lighting, five colors, pre-orders Oct 13, available Oct 30. Siri AI is not there at launch: it comes 'via a future software update'. Up to six household members recognized by voice
- Privacy claims (Apple): camera and microphone data on HomeView and microphone data on Apple TV are processed in Secure Exclaves, isolated from the system software, apps and Apple; Private Cloud Compute data 'is not stored or made accessible to Apple'
- Also announced: Apple Home Collection, smart-home accessories (cameras, thermostats, a doorbell, a lock) developed with LG Electronics and Schneider Electric, from December 2026, some in early 2027
- Executives quoted: Greg Joswiak (SVP Worldwide Marketing): HomeView 'is designed to be shared by everyone in the home yet feel completely personal to each user'; Bob Borchers (VP Worldwide Product Marketing) on Apple TV

#### What happened
Apple used press releases, not an event, to launch its long-rumoured home line: a 7-inch smart display called HomeView, a faster
Apple TV 4K with far-field microphones, and a second HomePod mini. The common selling point is Siri AI, the assistant Apple rebuilt
for iOS 27. On HomeView and Apple TV it is labelled beta; on HomePod mini 2 it arrives later.

The devices lean on the user's phone. Anyone can ask general questions, but requests that use personal data need a paired iPhone or
iPad that supports Apple Intelligence, and HomeView shows personal content only when it sees an enrolled face and that device is
nearby and recently unlocked. Apple says some features depend on server-side models with daily limits and that wider access will be
paid later.

HomeView is not on sale in the EU. Apple ties this to Siri AI being unavailable in iOS in the EU.

#### Why it matters
This is Apple's first new hardware category designed around its generative assistant, and the first Apple product withheld from the
EU as a whole because of an AI feature. It also shows Apple's deployment model: on-device and Private Cloud Compute inference, the
phone as the identity key, and a stated plan to charge for heavier assistant use.

Not verified here: reviews (WIRED, WSJ and The Verge were not read); how well Siri AI works on these devices.

Sources: [Apple Newsroom: Meet Apple HomeView, a new way to enjoy entertainment and manage the home](https://www.apple.com/newsroom/2026/10/meet-apple-homeview-a-new-way-to-enjoy-entertainment-and-manage-the-home/); [Apple Newsroom: Apple unveils the new Apple TV 4K, supercharged by Siri AI](https://www.apple.com/newsroom/2026/10/apple-unveils-the-new-apple-tv-4k/); [Apple Newsroom: Apple announces the next generation of HomePod mini](https://www.apple.com/newsroom/2026/10/apple-announces-the-next-generation-of-homepod-mini/); [Apple Newsroom: Introducing Apple Home Collection, a new lineup of smart home accessories](https://www.apple.com/newsroom/2026/10/introducing-apple-home-collection-a-new-lineup-of-smart-home-accessories/); [MacRumors: Apple HomeView won't be available in the EU](https://www.macrumors.com/2026/10/13/apple-homeview-wont-be-available-in-eu/); [Daring Fireball (John Gruber): HomeView and the registered personal device](https://daringfireball.net/linked/2026/10/13/homeview); [WIRED: Apple HomeView smart display (not read)](https://www.wired.com/story/apple-homeview-smart-display/); [The Verge: Apple TV 4K (4th gen) with Siri AI announced (not read; facts via Techmeme summary)](https://www.theverge.com/tech/1009663/apple-tv-4k-siri-ai-microphones-4th-gen-announced); [The Verge: Apple launches HomeView, its first smart display for the home (not read)](https://www.theverge.com/tech/1009704/apple-homeview-smart-display-first-look); [WSJ: My first impressions of Apple's new smart home display (paywalled, not read; Techmeme summary: square screen, personalized iPhone-like experience based on facial recognition for up to six users, Siri AI, closed ecosystem)](https://www.wsj.com/tech/personal-tech/my-first-impressions-of-apples-new-smart-home-display-615ac94c)

### 2026-10-13: Google Play adds a private 'personal library' for Android apps vibe-coded in AI Studio; no developer account needed

***; Google; Products; confidence high; https://postcutoff.com/e/2026-10-13-google-play-personal-vibe-coded-apps/

On Oct 13, 2026 Google said that Android apps generated in Google AI Studio can be launched on the creator's own devices through Google Play with a standard Google account, without a developer account and without a public store listing. The apps stay private; sharing with a small circle of friends and family, and a path to a regular developer account with full review, are due "in the coming months". The rollout starts in the US and select markets for users aged 18 and over. Google also expanded the Ask Play assistant (app comparisons), announced "Play in Gemini" and launched an AI game recommender, Game Finder.

- Google Play blog (Sam Bright, VP and GM, Google Play and Developer Ecosystem): 'Using Google AI Studio, you can vibe code a custom Android app ... through Google Play you can now launch these personal apps on your own device. This means your creations stay private, secure, and separate from the public store catalog. You only need your standard Google account to do so — no developer account required.'
- Availability: 'starts rolling out today to select markets, including the U.S., for users 18 and over'
- Coming 'in the coming months': sharing personal apps 'with a close circle of friends and family', and a way to 'graduate to a professional developer account', at which point the app goes through 'the same Trust & Safety reviews and quality standards as every professional developer'
- Before this, AI Studio's Android app builder (launched in May 2026) needed a phone connected to a computer over ADB to install an app (The Verge, 9to5Google). A 'Your personal library' section in the Play Store now lists the apps and restores them across the user's devices (9to5Google)
- Review: Ars Technica reports 'Private apps are not reviewed by Google' and that Google has not decided how many people count as a 'small group'; TechCrunch says the apps must still comply with Google Play's policies; 9to5Google says shared apps go through a security review. The accounts differ, and Google's post does not settle it
- Cost: ZDNet reports there is 'no charge to experiment and build an app' in AI Studio with a standard account
- Framing (Sam Bright, press briefing quoted by TechCrunch): Play is 'moving from there's an app for that to there's an app for me'
- Other AI features announced the same day: Ask Play can 'generate a comparison between apps, highlighting key differences and user sentiment'; 'Play in Gemini' brings apps, games and developer content into Gemini conversations and 'will start rolling out to all Gemini users soon'; Game Finder builds a 'Gamer DNA' profile for recommendations (US, English first)
- Scale figures differ: Google's post cites a 'community of over 2 billion players'; TechCrunch and Ars Technica cite 1.2 billion players for Game Finder's data

#### What happened
Google announced a set of Play Store changes on Oct 13, 2026, the same day it released Gemini 4 Argon. The main one concerns apps
that people generate for themselves. Since May 2026, Google AI Studio has been able to build native Android apps from prompts, but
getting one onto a phone meant connecting the phone to a computer, or publishing it to the whole Play Store through a developer
account. Now a creator can send the app from AI Studio to a private area of Google Play and install it on their own devices with an
ordinary Google account. The app does not appear in search or listings.

Two further steps are promised for the coming months: sharing an app with a small group of friends or family, and converting a
personal app into a public one through a developer account with the normal review.

Press accounts disagree on whether private apps are checked. Ars Technica says Google does not review them; TechCrunch says they
must comply with Play policies; 9to5Google says shared apps get a security review. Google's post says only that the apps are
"private, secure, and separate from the public store catalog".

#### Why it matters
- A major app store now distributes software that a model wrote for one user, outside the public catalog.
- It lowers the last practical barrier for non-developers on Android: installation. Ars Technica notes that Google makes this push
  "despite its slow progress in coding-focused AI models lately".
- Open points: how large the friends-and-family sharing group will be, and what review, if any, private and shared apps get
  (press accounts differ, see above).

Sources: [Google: New on Google Play: More ways to create, discover, and play (Sam Bright)](https://blog.google/products-and-platforms/platforms/google-play/google-play-updates-october-2026/); [TechCrunch: Google Play now lets anyone build private, vibe-coded apps that run on Android (Sarah Perez)](https://techcrunch.com/2026/10/13/google-play-now-supports-private-vibe-coded-apps-built-just-for-you/); [Ars Technica: Google wants to host your personal vibe-coded apps in the Play Store (Ryan Whitwam)](https://arstechnica.com/google/2026/10/google-wants-to-host-your-personal-vibe-coded-apps-in-the-play-store/); [The Verge: Google makes it easier to launch your vibe-coded apps (Emma Roth)](https://www.theverge.com/tech/1009066/google-play-vibe-coded-android-apps-launch); [9to5Google: Google Play adding 'personal library' to access your vibe-coded Android apps](https://9to5google.com/2026/10/13/google-play-vibe-coded/); [ZDNet: Google lets you make personal Android apps for free now, with just prompts](https://www.zdnet.com/tech/custom-android-apps-google-ai-studio/)

### 2026-10-13: StepFun's device unit launches the STEPX Neo 'LLM-native agent phone' at 3,999 yuan, bundled with 10 billion tokens a month for a year

***; StepFun; Products; confidence medium; https://postcutoff.com/e/2026-10-13-stepfun-stepx-neo-agent-phone-launch/

On Oct 13, 2026 in Shanghai, StepFun's device unit (阶跃终端) launched the STEPX Neo "early edition" (先行版), which it calls the first large-model-native agent phone. One version (16GB + 256GB) costs 3,999 yuan as a "full-stack price" covering the phone, the agent-native OS Step AOS 1.0, the Step Studio IDE and Step models, plus a "STEPX Token Plan" of 10 billion tokens a month for 12 months. Pre-sale opened at 21:00 the same day; sales start Oct 16 at 10:13. The built-in agent, Amoo, runs simple tasks on the device and sends complex and multi-app tasks to cloud models for planning. Chinese press reports list nearly 100 partner companies or products.

- Event: 'Ready Builder One', Shanghai, Oct 13, 2026 (Yesky via Sina Tech). Product name: STEPX Neo 先行版 (early edition); colors carbon black and silicon silver (MyDrivers)
- Price: 3,999 yuan for the only version, 16GB RAM + 256GB storage, described as a 'full-stack price' (全栈价) for the '软硬模' bundle: Step models, Step AOS, Step Studio and the phone (MyDrivers; IT Home via 17173)
- Dates: pre-sale Oct 13 at 21:00; open sale Oct 16 at 10:13 (Beijing time)
- STEPX Token Plan: buyers get a quota of 10 billion tokens (百亿 Token) a month for 12 consecutive months. Which models the quota covers and what happens after 12 months were not stated in the reports read
- Chipset: reports say 'Snapdragon 8 series' without naming the model (IT Home). A Weibo tech blogger (WHYLAB) wrote Snapdragon 8 Gen 5; this is unconfirmed. A 6.36-inch screen appears in user hands-on posts, not in an official spec sheet we read
- Step AOS 1.0: an agent-native operating system with long-term memory, task scheduling and safety mechanisms at the system layer; 'dual-domain memory' (a user domain for personal information and preferences, an agent domain for knowledge and methods); principles stated as trusted, visible, controllable, reversible (可信、可见、可控、可逆)
- Device vs cloud: quick tasks such as alarms and system settings run on the device, some offline; for information queries and single-app or multi-app tasks a cloud model plans and splits the task and the device executes it. The launch reports do not name the on-device model; StepFun's July announcement called it StepEdge. IT Home names Step 5 Preview as a Step model (in connection with Tencent's WorkBuddy)
- Amoo (阶跃Amoo) agent: can call system-level functions, use Alipay life services with the user's authorization, work across apps, and show its status on the rear MiniLED dot-matrix sub-screen; users can manage its permissions and memory and see task progress
- Partners: 'nearly 100 companies or products', including Alipay, Tencent, Ctrip, Didi, Jianying (CapCut), Amap and Meituan. Alipay: system-level agent cooperation and customised AI payment; STEPX is described as the first device maker to join Alipay Vibe Pay. Tencent: WorkBuddy, WeCom and WeRead at launch. Jianying connects through the A2A protocol
- Developers: Step Studio, an 'AI-native IDE' that builds apps, games, agents and Skills from natural-language descriptions; the system exposes sensors, camera and the rear screen and supports MCP and A2A
- Hardware extras: six-microphone array, a low-power audio-sensing chip and a pickup algorithm called Whisper 1.0 (name as reported; not OpenAI's Whisper as far as the reports say); transparent, 'retro-future' design
- Amoo Growth Plan: the first 10,000 users who meet task targets get a 'Star Ticket' (星辰船票) to swap the Neo for any new STEPX device within five years at no cost
- Background: StepFun announced the STEPX brand, StepAOS, Amoo, the on-device model StepEdge and the Neo on July 13, 2026 in Shanghai, where chairman Yin Qi presented the phone (Xinhua, July 14; the Xinhua page is marked as promotional content). No price or date was given then
- Not verified: sales numbers; the national-standard 'L3' AI-terminal grading mentioned in pre-launch reports does not appear in the launch reports read; no executive quotes from the Oct 13 event were found

#### What happened
StepFun (阶跃星辰), a Shanghai model lab, set up a device unit and in July 2026 announced a hardware brand, STEPX, with an operating
system built around agents instead of apps. On Oct 13 the first product went on sale: the STEPX Neo early edition. The company prices
it as one package of phone, operating system, developer tool and model access, at 3,999 yuan, and adds a year of model usage (10 billion
tokens a month).

The phone's agent, Amoo, is part of the system. Simple commands run locally. For larger requests a cloud model plans the steps and the
phone carries them out across apps, with partners such as Alipay, Tencent's WorkBuddy and WeCom, Ctrip, Didi, Meituan, Amap and
Jianying exposing their services to the agent through protocols such as MCP and A2A. A small rear screen shows what the agent is doing.

#### Why it matters
Most "AI phones" add an assistant to Android. This is a model lab selling its own handset where the agent is the main interface and the
token allowance is part of the price, which tests whether model companies can reach users through hardware. It arrives in the same week
as Ant's LingGuang and Alibaba's Qwen Go personal-agent betas.

Limits of this entry: it is written from Chinese tech-press reports of the launch event; StepFun's own launch page was not found. The
exact chipset, the name of the on-device model in the shipping phone, and the terms of the token plan are not confirmed.

Sources: [Sina Tech / Yesky (Oct 13, 2026): 阶跃终端正式发布STEPX Neo：主打跨应用执行任务 售价3999起](https://finance.sina.com.cn/tech/roll/2026-10-13/doc-inivawmy0546861.shtml); [MyDrivers (Oct 13, 2026): 首款大模型原生智能体手机！阶跃STEPX Neo先行版发布：3999元](https://news.mydrivers.com/1/1157/1157185.htm); [IT Home via 17173 (Oct 13, 2026): STEPX Neo 先行版发布：骁龙 8 系芯片，全栈价 3999 元起](https://news.17173.com/content/10132026/220332641.shtml); [Weibo via Sina (WHYLAB, Oct 13, 2026): launch post with chipset claim](https://www.sina.cn/weibo/detail/5353652253623413.html); [Xinhua (July 14, 2026): 阶跃星辰推出 STEPX 终端品牌 首款大模型原生智能体手机同期亮相](https://www.news.cn/digital/20260714/8354f82874ee4741ad1fd9ef716112ef/c.html); [Sina Tech / IT Home (Oct 8, 2026): launch set for Oct 13 at 19:00](https://finance.sina.cn/stock/jdts/2026-10-08/detail-iniunrri4395249.d.html?vt=4)

### 2026-10-13: Alibaba launches Qwen N1 and N1 Pro AI glasses in China from RMB 2,899, with batteries swappable while worn

**; Alibaba; Products; confidence medium; https://postcutoff.com/e/2026-10-13-alibaba-qwen-n1-ai-glasses/

On Oct 13, 2026 Alibaba launched its Qwen N1 AI glasses in China at RMB 2,899 and the N1 Pro at RMB 3,399. Both run Qwen-powered features (translation, AI memos, automatic meeting notes, travel reminders) with a 50-megapixel camera and 4K video, and carry replaceable batteries in the temples that can be swapped without taking the glasses off; the Pro adds eye-tracking control, iris verification for payments and heart-rate, blood-oxygen and skin-temperature monitoring.

- Prices: Qwen N1 from RMB 2,899; N1 Pro from RMB 3,399 (standard packages)
- Battery: replaceable batteries in the temples, swappable during use; claimed 9 hours (N1) and 8 hours (N1 Pro)
- Camera: 50 MP with 4K video recording
- AI features: translation, AI memos, automatic meeting notes and travel reminders, powered by Qwen (which Qwen version is not stated)
- N1 Pro additions: eye-tracking controls, iris verification for payments, heart rate, blood oxygen and skin temperature
- Not stated in the report: display details, chipset, the Qwen model version, and shipping dates beyond the launch

#### What happened
Alibaba put its assistant on a face-worn device, following Chinese rivals and Meta into AI glasses. The specifications here
come from TechNode's write-up of a Chinese IT Home report; Alibaba's own announcement page was not located, so confidence is
medium and the figures are vendor claims relayed at second hand.

#### Why it matters
AI glasses are becoming the main consumer hardware fight in China, and the swappable battery is a direct answer to the
battery-life complaints that dog Meta's Ray-Bans. It also puts Qwen in front of users continuously, which matters for
Alibaba's push to make Qwen a personal agent.

Sources: [TechNode: Alibaba launches Qwen N1 AI glasses with swappable batteries, starting at RMB 2,899 (Oct 14)](https://technode.com/2026/10/14/alibaba-launches-qwen-n1-ai-glasses-with-swappable-batteries-starting-at-rmb-2899/)

### 2026-10-13: Meta brings 'Your Algorithm' to Facebook: users describe in plain language what they want in Feed and Reels

**; Meta; Products; confidence medium; https://postcutoff.com/e/2026-10-13-facebook-your-algorithm-ai-feed-controls/

On Oct 13, 2026 Meta extended its AI feed-customisation tool "Your Algorithm" from Instagram to Facebook's Feed and Reels. Users describe in plain language what they want to see more or less of; the system may ask follow-up questions and changes recommendations in real time. The Verge reports that Meta also said it is using large language models more in its ranking systems.

- Feature: 'Your Algorithm' on Facebook Feed and Reels, reached from the three-dot menu, with 'Interested' / 'Not interested' buttons on posts (TechCrunch)
- How it works: the user writes preferences in natural language; the tool can ask follow-up questions about what they liked or disliked; changes show in recommendations in real time
- History: launched on Instagram Reels in December 2025, extended to Instagram Feed and Explore in June 2026; Threads has a similar feature, 'Dear Algo' (TechCrunch)
- Techmeme's summary of The Verge (article not accessible to us): Meta 'says it is using LLMs more in its ranking systems'
- Not stated in the reports read: which model interprets the requests, and the rollout countries and schedule

#### What happened
Facebook users can now tell the recommendation system what they want in words, as Instagram users already could. Meta presents it
as giving people more control over the feed.

#### Why it matters
Language models are moving into the ranking systems of the largest social networks, both as an interface for user control and,
by Meta's account, inside the ranking itself.

Not verified: Meta's own announcement was not found; details come from TechCrunch and a one-line summary of The Verge.

Sources: [TechCrunch: Facebook is giving users more control over their algorithm (Lauren Forristal)](https://techcrunch.com/2026/10/13/facebook-is-giving-users-more-control-over-their-algorithm/); [The Verge: Facebook will let you customize your feed algorithm using AI (not read; via Techmeme summary)](https://www.theverge.com/tech/1010344/meta-facebook-your-algorithm-ai)

### 2026-10-14: Apple plans to launch Apple Intelligence in mainland China in December with iOS 27.3, using Alibaba's Qwen models; no new Siri (Bloomberg)

***; Apple, Alibaba, Baidu; Products; confidence medium; https://postcutoff.com/e/2026-10-14-apple-intelligence-china-december-alibaba/

Bloomberg reported on Oct 14, 2026 that Apple will bring Apple Intelligence to mainland China in December through iOS 27.3, more than two years after the features launched elsewhere. The China version will use Alibaba's Qwen and Tongyi Wanxiang models after permits granted on Sept 29. It will start with writing tools, notification summaries, Genmoji and Clean Up, without the redesigned Siri or Image Playground.

- Timing: December 2026 via iOS 27.3 (Bloomberg, via TechNode)
- Approval: Apple received permits on Sept 29, 2026 allowing recent iPhones to support Apple Intelligence and the Alibaba models
- Models: Alibaba's Qwen and Tongyi Wanxiang; Apple is also working with Baidu on search and Siri-related functions
- Initial features: writing tools, notification summaries, Genmoji, Clean Up; excluded: the redesigned Siri AI assistant and Image Playground
- Features are those first rolled out elsewhere in 2024 (Techmeme summary of Bloomberg)
- Earlier steps: on July 15, 2026 the Cyberspace Administration of China added Apple's generative AI to its list of approved providers, and Alibaba and Baidu confirmed their roles, with no launch date (Bloomberg via The Star). TechNode's Sept 29 'permits' date may refer to a later device-level approval; the two reports were not reconciled
- Same week: Apple CEO John Ternus told Time 'all options should be on the table' after reports that Apple tested Chinese chips (Techmeme)
- False start: on March 30, 2026 users on X and Weibo reported Apple Intelligence going live in China (iOS 26.4 apparently needed); Mark Gurman said it 'was a mistake and it has since been pulled' and that Apple was still awaiting regulatory approval though the features had been ready for months (9to5Mac update)
- Tim Cook in Shanghai (interview with The Paper, Oct 15, 2026; via TechNode Oct 16): the first Apple Intelligence features and the associated Alibaba Qwen models have regulatory approval; a beta comes later in October with live translation, writing tools and photo cleanup; the redesigned Siri AI is not in the first batch. Reuters headline: 'Tim Cook says Apple Intelligence approved in China'

#### What happened
China is Apple's largest market without Apple Intelligence. Foreign AI models need Chinese regulatory approval there, so Apple
partnered with Alibaba. China listed Apple's generative AI as approved on July 15; TechNode, summarizing Bloomberg, says further permits came on Sept 29, and the first features ship in December with iOS 27.3. The
new Siri is not included.

#### Why it matters
The launch would put Alibaba's Qwen models on a large base of iPhones in China. It also shows the split AI stack that global
consumer companies now run: US models outside China and approved Chinese models inside.

Unverified: Bloomberg's article is paywalled; details come from TechNode's summary and Techmeme's headline.

Sources: [Bloomberg: Apple prepares to debut first AI features in China in December](https://www.bloomberg.com/news/articles/2026-10-14/apple-prepares-to-debut-first-ai-features-in-china-in-december); [TechNode: Apple reportedly plans December China launch for Apple Intelligence using Alibaba models](https://technode.com/2026/10/15/apple-reportedly-plans-december-china-launch-for-apple-intelligence-using-alibaba-models/); [The Star (Bloomberg, Jul 16): Apple gets approval for iPhone AI in China with Alibaba, Baidu](https://www.thestar.com.my/tech/tech-news/2026/07/16/apple-gets-approval-for-iphone-ai-in-china-with-alibaba-baidu); [Time: John Ternus on the iPhone Duo, memory shortages and AI](https://time.com/collection/best-inventions/2026/apple-iphone-duo/); [9to5Mac (Mar 30): Apple Intelligence rolling out in China per user reports (updated: a mistake, pulled, per Gurman)](https://9to5mac.com/2026/03/30/apple-intelligence-rolling-out-now-in-china-per-user-reports/); [TechNode: Apple Intelligence to enter China beta testing this month, Tim Cook confirms (Oct 16)](https://technode.com/2026/10/16/apple-intelligence-to-enter-china-beta-testing-this-month-tim-cook-confirms/); [Reuters: Tim Cook says Apple Intelligence approved in China, paper reports (Oct 15)](https://www.reuters.com/business/retail-consumer/tim-cook-says-apple-intelligence-approved-china-paper-reports-2026-10-15/)

### 2026-10-14: Google opens Deep Review, its agentic document-verification tool piloted on 16,000+ conference papers, to the public at Research@ 2026

***; Google Research, Google DeepMind; Products; confidence high; https://postcutoff.com/e/2026-10-14-google-deep-review-public-launch-research-2026/

At its Research@ 2026 event on Oct 14, 2026 Google Research launched Deep Review publicly. It is an agentic system that checks long technical documents section by section and gives verification feedback. It grew out of the Paper Assistant Tool (PAT), which gave optional pre-submission feedback on more than 16,000 manuscripts at STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027. Anyone can now use it for two reviews a week free (more with a Gemini API key), on papers as well as legal, financial and design documents. It now supports Gemini 4 Argon.

- Public worldwide launch on Oct 14, 2026; free tier: 2 document reviews per week per user; higher volume and API access with a Gemini API key; PDFs and plain text first, multi-file 'coming soon'
- Pilots: 16,000+ opt-in pre-submission reviews at STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027, kept separate from official peer review
- Pipeline: segments the document, budgets compute by information density, runs parallel specialist verification agents with full-document context, then a global synthesis with cross-section consistency checks and external grounding
- Google: 'Documents uploaded to Deep Review are never used to train, fine-tune, or improve Google models'; uploads kept one year for display
- Context given by Google: NeurIPS/ICML/ICLR submissions topped 74,000 in 2026, vs ~45,000 (2025), 32,000 (2024), 23,000 (2023)
- Also an Antigravity 'Teamwork Deep Document Review Pattern' (/teamwork-preview); authors Rajesh Jayaram and Vincent Cohen-Addad
- Other Research@ 2026 items (Yossi Matias): Earth AI early-access expansion with an experimental Geospatial Reasoning agent API and $10M from Google.org; TimesFM and TabFM for all BigQuery users; MedGemma >10M downloads and a Nature Medicine paper; Groundsource dataset of 2.6M+ flood events; Synaptics edge-AI chip on Google's open IP to be shown 'next week'; Play with Putty prototyping in Google Labs

#### What happened
Google Research held its annual Research@ event on Oct 14, 2026. The main product news was the public launch of **Deep Review**, an agentic
system "engineered to provide rigorous feedback and technical verification on complex documents". It began as the Paper Assistant Tool
(PAT), which authors could use for private feedback before submitting to STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027; Google says the
pilots covered more than 16,000 manuscripts. The public version also handles legal agreements, financial reports, design docs and
presentations, and has Gemini 4 Argon support. The blog post does not give accuracy figures in text (they are only in a table image).

The same day, Yossi Matias's Research@ roundup listed nine other items, among them an experimental Geospatial Reasoning agent API in Earth AI,
TimesFM and TabFM for BigQuery users, MedGemma's Nature Medicine paper and the AMIE Lancet study (already covered in this dataset).

#### Why it matters
Conference submissions have tripled since 2023 and reviewers are overwhelmed (Google cites more than 74,000 submissions to the three big ML
conferences in 2026). A free, general-purpose AI verifier from a major lab makes automated pre-review routine for authors and anyone checking
long documents. It also raises the question of whether reviewers and authors will rely on the same model family to check each other's work.

Sources: [Google Research: Analyzing technical documents with Deep Review (Oct 14)](https://research.google/blog/analyzing-technical-documents-with-deep-review/); [Google Research: Research@ 2026: 10 ways we're amplifying human ingenuity (Yossi Matias, Oct 14)](https://research.google/blog/research-2026-10-ways-were-amplifying-human-ingenuity/); [Deep Review (try it)](https://research.google/deepreview); [blog.google: Research@ 2026: Novel capabilities and intelligence](https://blog.google/innovation-and-ai/technology/research/research-capabilities-intelligence-2026/); [arXiv 2606.28277: Towards Automating Scientific Review with Google's Paper Assistant Tool](https://arxiv.org/abs/2606.28277); [Google Research: Putting Earth AI to work: geospatial agents (Oct 14)](https://research.google/blog/putting-earth-ai-to-work-geospatial-agents-for-our-planets-biggest-challenges/)

### 2026-10-14: OpenAI deprecates GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini in the API: shutdown January 14, 2027, replacement gpt-6-luna

***; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-10-14-openai-deprecates-gpt-4o-gpt-4-1-api/

On 2026-10-14 OpenAI told developers that the remaining GPT-4-series API models (gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14) will be removed from the API on 2027-01-14, with gpt-6-luna as the recommended replacement. Together with an Oct 1 notice (gpt-5.3-codex, gpt-5.1, gpt-5.4-nano and the tts-1 / gpt-4o-mini-tts voices), it ends the API life of the GPT-4 generation.

- Notice date: 2026-10-14; shutdown date: 2027-01-14 (three months' notice)
- Models: gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14, gpt-4.1-mini-2025-04-14
- Recommended replacement for all five: gpt-6-luna
- Fine-tuned versions of these models follow the same shutdown schedule as their base models
- Earlier: gpt-4o-2024-05-13 and gpt-4.1-nano were already scheduled for shutdown on 2026-10-23
- Oct 1, 2026 notice: gpt-5.3-codex and gpt-5.1 removed 2027-04-01 (replacement gpt-6-sol), gpt-5.4-nano removed 2027-04-01 (replacement gpt-6-luna)
- Oct 1, 2026 notice: tts-1, tts-1-hd, gpt-4o-mini-tts-2025-03-20 and gpt-4o-mini-tts-2025-12-15 removed 2027-01-06 (replacement gpt-realtime-2.1-mini)

#### What happened
OpenAI's API deprecations page carries a section dated 2026-10-14 ("GPT-4 series model deprecations"): developers using
the listed GPT-4o and GPT-4.1 snapshots were notified that the models will be removed from the API on January 14, 2027.
For every model the page names `gpt-6-luna`, OpenAI's small, cheap GPT-6 tier, as the replacement. Fine-tuned models
built on these bases are shut down on the same date, which matters because GPT-4o and GPT-4.1 were among the last
fine-tunable OpenAI chat models.

Two weeks earlier, a 2026-10-01 section had scheduled `gpt-5.3-codex`, `gpt-5.1` and `gpt-5.4-nano` for removal on
April 1, 2027 (six months' notice) and the older text-to-speech models (`tts-1`, `tts-1-hd` and both `gpt-4o-mini-tts`
snapshots) for January 6, 2027, pointing voice users to `gpt-realtime-2.1-mini`.

The notice was found on the deprecations page itself (and through the aiweekly.co news list of Oct 15); no OpenAI blog
post or press coverage of it had been found as of 2026-10-15. Whether ChatGPT or Azure OpenAI are affected is not stated
on the page.

#### Why it matters
GPT-4o (May 2024) was OpenAI's first natively multimodal "omni" model and the base of its audio/realtime variants; GPT-4.1
(April 2025) was the 1M-token non-reasoning workhorse many production apps still pin. Pointing all of them to gpt-6-luna
shows OpenAI consolidating the API on the GPT-6 family, and gives developers with fine-tunes three months to migrate.

Sources: [OpenAI API deprecations page (2026-10-14 and 2026-10-01 sections)](https://developers.openai.com/api/docs/deprecations); [GPT-4o model page](https://developers.openai.com/api/docs/models/gpt-4o)

### 2026-10-14: Epic releases a 'vibe coded' Preview of a rewritten Epic Games Launcher, saying it is faster than Steam

**; Epic Games; Products; confidence high; https://postcutoff.com/e/2026-10-14-epic-games-launcher-vibe-coded-preview/

On Oct 14, 2026 Epic Games released a public "Preview" of a rewritten Epic Games Launcher, announcing: "programming is hard. So, instead of doing that, we vibe coded a new Epic Games Launcher that's way faster and more responsive than the current one". Store GM Martin Keely said a small team used AI to write and review the new code from scratch in a different tech stack, for Windows, Mac and later Linux. Ars Technica's test found it reached the store page a few seconds faster than Steam from a cold start; Epic warns of bugs.

- Epic's announcement: 'we vibe coded a new Epic Games Launcher that's way faster and more responsive than the current one, and has a new look'; on X: 'We vibe coded the Epic Games Launcher to be faster than the other guy'
- Martin Keely (VP, GM of the Epic Games Store) to PC Gamer: the team 'started from scratch' and used AI as 'a tool to write and review new features and new capabilities from the ground up in a different tech stack'; to IGN: AI 'allowed a small team to rewrite the code base and be able to launch it faster across platforms', with Windows, Mac and a planned Linux version
- Ars Technica's own test: cold start to the store page a few seconds faster than Steam; Epic says the Preview 'will have some bugs'

#### What happened
Epic rebuilt its PC game launcher with AI coding tools and released it as an opt-in Preview, marketing the AI-written rewrite openly rather
than hiding it.

#### Why it matters
A large consumer app from a major games company is now advertised as "vibe coded", a sign of how quickly AI-written production software has
become a selling point. Which AI tools were used was not stated.

Sources: [Ars Technica: Epic says its new vibe-coded Epic Games Launcher is faster than Steam (Oct 14)](https://arstechnica.com/gaming/2026/10/programming-is-hard-epic-games-leans-on-vibe-coding-for-its-new-launcher-app/)

### 2026-10-15: Tesla unveils the next-gen Roadster with SpaceX-derived cold-gas thrusters; driverless stunt cars leave an over-vertical ramp and 'kickflip' over parked Teslas

***; Tesla, SpaceX; Products; confidence high; https://postcutoff.com/e/2026-10-15-tesla-roadster-unveil-thrusters-kickflip/

On the evening of Oct 15, 2026 (US Central time; Oct 16 UTC) Tesla livestreamed the unveiling of its long-delayed second-generation Roadster from an event outside Waco, Texas (Reuters: Musk spoke at SpaceX's McGregor test site). The car has a fan "suction" system and, on its top trims, SpaceX-derived cold-gas thrusters. In two stunts with nobody in the car, one car drove up a ramp that curves past vertical, left it upside down, righted itself on gas jets and landed in a cloud of gas; another lifted off flat ground on thrusters and did a full roll ("kickflip") over a parked Semi, Cybertruck, Cybercab, Model Y and Model 3. Tesla gave three trims (Performance, Thrust, Rocket), 0–60 mph claims under 1.6 / 0.9 / 0.59 s, 300+ miles of range, production in 2028, and no prices. Clips of the stunts got millions of views. Whether the flights had help that can't be seen (cables, a lighter stunt body) has not been established; press reports call them real, staged stunts.

- Livestream on Oct 15, 2026, scheduled for 7:55 pm Central (00:55 UTC Oct 16) after an Oct 1 date was cancelled for weather; replay 'Roadster Event' on Tesla's YouTube channel (23:21, 1.7M views in ~6 h) (Electrek, Tesla)
- Venue: an event outside Waco, Texas (TechCrunch); Reuters says Musk spoke at SpaceX's test site in McGregor, Texas
- Trims (Tesla's X thread, via Electrek): Performance (2 seats, active aero, 'multi-mode suction system', 0–60 under 1.6 s), Thrust (2 seats, rear cold-gas thrusters, 'race ready', under 0.9 s, limited release), Rocket (1 seat, perimeter thrusters for vertical and rotational movement, 'air ready', under 0.59 s, over 230 mph, invite only)
- Tesla's Roadster page says the Rocket is not approved for public road use (Electrek, New Atlas)
- Range over 300 miles; 2017 promise was 620 miles, 1.9 s 0–60, 250+ mph and a 2+2 layout (Electrek)
- Lars Moravy: 10 thrusters fed by 600 litres of rechargeable compressed air; every usable void filled with carbon-fibre-overwrapped pressure vessels (COPVs) (TechCrunch)
- Tesla claims the thrusters give four to five times the force the tyres can transmit; rear thrusters over 80 kN; fan system holds the car upside down even when stationary; 2 g cornering (Electrek, New Atlas)
- Tesla's YouTube description: 'cold-gas thrusters w/ all-electric actuators make micro adjustments within a fraction of a second to stabilize the vehicle in flight'
- Live 0–60 run: 0.81 s after timer problems; recorded run 1.0 s (Electrek)
- Reservations: $5,000 refundable deposit plus $45,000 within 10 days for Performance ($50k total); Thrust $250k deposit; Rocket invite only; no prices announced (Electrek, Reuters, New Atlas)
- Production: 2028 (Musk: '2028, actually'); first shown Nov 2017 with 2020 deliveries promised (Reuters, Electrek)
- Stunt 1 (ramp demo, stream 6:03–10:55): car climbs an over-vertical curved ramp, leaves it inverted, rights itself on gas jets and lands in a gas cloud; Tesla's on-screen HUD read ~66.7 ft at the top
- Stunt 2 (kickflip, stream 10:56–12:43): Tesla calls it 'the 2nd live demo during the Roadster event'; its 44-s edit 'The Kickflip' / 'Do a kickflip!' got 5.0M views on X in about a day
- Nobody was in the stunt cars (Electrek); pre-event, The Information reported the demo car would be 'operated remotely with nobody in it'; Tarmac Life reports Tesla said it ran the car autonomously 'to make sure we don't kill anyone' and that a Starman dummy sat in the seat
- Musk on stream: 'a friend of mine, Peter Thiel, once said that the future promised us flying cars, but instead we got 140 characters. Well, not anymore.'
- Collin Rugg's clip of the stream: 7.17M views; Tesla's kickflip post: 5.01M; unusual_whales' 7-s clip: 1.06M (fxtwitter, Oct 16)

#### What happened
Tesla first showed a second-generation Roadster prototype in November 2017 and promised deliveries in 2020. After at least six production
slips and seven unveil dates (Electrek's count), and an Oct 1 event in Texas cancelled for weather, Tesla livestreamed the reveal on the
evening of Oct 15, 2026 (scheduled 7:55 pm Central). Musk had promised the "most exciting, whether it works or not, demo ever of any
product" (TechCrunch). Electrek and Reuters tie the presentation to SpaceX: the thrusters, pressure vessels and fans are presented as SpaceX
technology.

**The car.** Tesla's design language "from S3XY to Cybertruck, Cybercab & Optimus", inspired by the SR-71 and F-35: hidden headlights,
a full-length taillight, "Ano White" paint with rocket-coating properties, gold-tinted glass like a spacesuit visor, a steer-by-wire yoke,
a fixed seat and manual controls to save weight, and magnesium, titanium and carbon fibre (Tesla, Electrek). A "multi-mode suction
system" of fans pulls air from under the car hard enough to hold it upside down; in Track Mode active wings, driven by actuators from the
Optimus programme, add downforce (Electrek). Three trims: **Performance** ("road ready"), **Thrust** (rear cold-gas thrusters, "race
ready") and **Rocket** (single seat, 10 cold-gas thrusters for vertical and rotational movement, "air ready", invite only, not approved
for public roads). Musk said production starts in 2028 and called the car, in TechCrunch's paraphrase, "the best of the last of the
human-driven cars". Musk said he might try to jump the Snake River Canyon (TechCrunch).

**The stunts.**
1. *Ramp demo* (replay 6:03–10:55). The car drives up a curved ramp that bends past vertical, sits upside down at the top on its suction
   system, leaves the ramp inverted, rights itself as gas jets fire, and lands in a large cloud of gas before driving away. Tesla's
   on-screen HUD showed G-force and altitude (about 66–67 ft at the top). After the landing Musk says "Well, it worked," and later the
   Peter Thiel line about flying cars and "140 characters" ("Well, not anymore").
2. *Kickflip* (replay 10:56–12:43). On flat ground beside a parked line-up (Semi, Cybertruck, Cybercab, Model Y, Model 3), a car lifts
   off on downward jets, makes a full roll over the Semi and lands in gas. Tesla's YouTube description calls it "the 2nd live demo during
   the Roadster event"; Tesla's 44-second edit was posted on X as "Do a kickflip!".

Electrek reports that nobody was in the stunt cars, that different cars were used for the demos, and that spectators were kept far away.
In August, The Information (via Electrek) reported that the demo car would be "operated remotely with nobody in it" and that an earlier,
more ambitious plan (magnetized roller-coaster ramp, upside-down driving, hover) had been scaled back: "We started off thinking we were
almost making an aircraft and ended up with something closer to hovering." After the event, Tarmac Life reported that Tesla said the car
was run autonomously "to make sure we don't kill anyone" and that a Starman dummy sat in the seat. Whether the cars were remotely piloted,
ran pre-programmed/automated control, or both is not clear from the sources. Tesla says the thrusters' all-electric actuators make
"micro adjustments within a fraction of a second to stabilize the vehicle in flight"; no source we found says AI models (Grok or others)
controlled the flight.

**Was it real?** No outlet we read reports evidence of cables, cranes or CGI, and none has independently verified that the flights were
unassisted. Reuters and TechCrunch describe the stunts as shown, without questioning them. Electrek (Jameson Dow) says it treats the demo
"as if it was real, not faked", calls fan cars and thruster stunts plausible, doubts the 4,700 kg downforce figure shown on Tesla's
display ("some liberties were taken"), and speculates that the demo car may not have had a normal battery pack. New Atlas says the
"flying" means "big jumps, barrel rolls, and quick moments of hovering", not sustained flight, and that it "can't be sure if there was a
human driver". AutoNotion (opinion) argues the stunts were shown as video segments and were "clearly scripted", not real-time; Tesla's own
description calls the kickflip a live demo, and an attendee's YouTube video is titled "I Went To The Tesla Roadster Unveil! Insane Stunts…" (Out of Spec Reviews,
https://www.youtube.com/watch?v=7oFL2ObpM08; not reviewed in detail).
Marques Brownlee asked "This is a prank, right?" (quoted by TechCrunch). AutoNotion also reports Musk saying "We need an audience to prove
it's not AI"; we have not found that line in another source.

Two AI video models and Claude also examined the clips (see the video files): they found that the car stays in the air longer than free
fall would allow, consistent with the claimed thrust, but footage at night cannot rule out hidden support. These are model observations,
not facts.

#### Why it matters
Tesla is increasingly valued as an AI and robotics company (Reuters notes it stopped selling Model S and X this year to free production
for Optimus robots and launched the Cybercab), and Musk presented the Roadster as a last human-driven car. The reveal is here mainly
because its stunt clips spread far beyond car media, look unreal, and are the kind of footage that later models will be asked about:
the car is a real announced product, the stunts were staged with driverless cars, and the flying Rocket trim is invite-only and not
road-legal.

Sources: [Tesla: Roadster Event (livestream replay, YouTube)](https://www.youtube.com/watch?v=qt5p-_tE8d0); [Tesla: The Kickflip | Roadster (YouTube)](https://www.youtube.com/watch?v=r5vsHHPVhJg); [Tesla: Roadster | The Last Best Driver's Car (YouTube, spec description)](https://www.youtube.com/watch?v=cvKIR-c1SsU); [Tesla Roadster page](https://www.tesla.com/roadster); [Tesla on X: 'Do a kickflip!'](https://x.com/Tesla/status/2110926164306931860); [Reuters (via Spokesman-Review): Tesla unveils next generation Roadster sports EV with SpaceX thrusters](https://www.spokesman.com/stories/2026/oct/15/tesla-unveils-next-generation-roadster-sports-ev-w/); [TechCrunch: Tesla reveals SpaceX-powered Roadster that flies and falls with style](https://techcrunch.com/2026/10/15/tesla-reveals-spacex-powered-roadster-that-flies-and-falls-with-style/); [Electrek: Tesla unveils Cyberfied new Roadster that can kickflip](https://electrek.co/2026/10/15/tesla-unveils-cyberfied-new-roadster-kickflip/); [Electrek: Are Tesla's claims for its new 'flying' Roadster real, or is it just full of cold gas?](https://electrek.co/2026/10/15/an-updated-technical-look-at-teslas-kick-flipping-0-60-in-1s-next-gen-roadster/); [New Atlas: Do the Tesla Roadster's aerial stunts count as flying?](https://newatlas.com/automotive/tesla-roadster-flying-ev-speed-thrusters-spacex/); [CNBC: Tesla reveals next-generation Roadster after almost a decade of delays](https://www.cnbc.com/2026/10/15/tesla-roadster-reveal-what-to-know-about-the-delayed-ev.html); [Tarmac Life: I watched the Tesla Roadster demo so you don't have to](https://www.tarmaclife.co.nz/news/i-watched-the-tesla-roadster-demo-so-you-dont-have-to/); [AutoNotion: The Tesla Roadster reveal wasn't a live demo (opinion)](https://www.autonocion.com/us/tesla-roadster-live-demo-recap/); [Electrek (Aug 14): Tesla 'flying' Roadster demo could come this month, with no one inside (citing The Information)](https://electrek.co/2026/08/14/tesla-flying-roadster-demo-this-month-no-driver/); [Electrek (Oct 13): Tesla sets Roadster event for Thursday after nearly 9 years of delays](https://electrek.co/2026/10/13/tesla-roadster-event-thursday-nine-years-delays/); [Collin Rugg on X: clip of the ramp stunt (7.1M views)](https://x.com/CollinRugg/status/2110901393238474863); [unusual_whales on X: 7-second kickflip clip](https://x.com/unusual_whales/status/2110910422358561071)

### 2026-10-15: Amazon launches Bee Frontier ($75) and Bee Eclipse ($100), always-available AI recorders that summarize conversations, with a $12/month Bee Pro plan

**; Amazon; Products; confidence high; https://postcutoff.com/e/2026-10-15-amazon-bee-frontier-eclipse-ai-wearables/

On Oct 15, 2026 Amazon's Bee unit launched two "personal AI" devices that listen to conversations and turn them into summaries, to-dos and memories. Bee Frontier is a $74.99 wristband or clip, and Bee Eclipse is a $99.99 card that sticks to the back of a phone. A Bee Pro subscription ($11.99/month) adds Alexa+, Amazon shopping and third-party integrations. Amazon says audio is processed in real time and deleted, never stored.

- Bee Frontier: $74.99, available now; wrist or clip; IP67; up to 7 days battery; dual mics; mic button, indicator light and auto-off after silence
- Bee Eclipse: from $99.99, 'available soon'; credit-card-size, magnetic phone attachment; 21+ days battery; call participants are prompted for permission each time recording starts
- Bee Pro: $11.99/month (30 days free): unlimited summaries, Amazon.com and Alexa+ integrations, third-party apps, to-dos, daily memories, chat assistant, desktop app; free tier has summaries and 14-day memory
- Privacy claims: audio processed into insights in real time then deleted; 'Bee Private Compute' that Amazon says it cannot access; transcripts encrypted with device-held keys
- Amazon's page does not name the AI model used
- Devices chief Panos Panay told TechCrunch he sees AI devices replacing the need to use apps
- Reach on X (fxtwitter, Oct 16 evening): Andy Jassy's launch post ~417k views and 213 replies ('audio is never stored, and only you can see your data'); @bee__computer launch ~134k; Panos Panay ~52k; @amazon ~49k; Mark Gurman's Bloomberg summary ~48k. David Pawlan (~36k) linked the Amazon.com shopping integration to Amazon blocking outside AI agents: 'It looks like there may be a reason Amazon is blocking AI agents after all'

#### What happened
Amazon, which bought the Bee wearable startup, launched two redesigned recorders. Both summarize what the wearer hears and connect
to Alexa+ and Amazon shopping through a paid plan. The Verge noted a day earlier that the Bee assistant can now also help users buy
things.

#### Why it matters
Big tech is moving into ambient, always-listening AI hardware. That raises consent questions for the people being recorded, which
Amazon addresses with indicator lights and call prompts.

Sources: [Andy Jassy's launch post on X (~417k views)](https://x.com/ajassy/status/2110729551571456391); [About Amazon: Amazon introduces 2 new AI products from Bee](https://www.aboutamazon.com/news/devices/amazon-personal-ai-bee-frontier-bee-eclipse); [TechCrunch: Amazon's Bee launches two new personal AI devices, starting at $75](https://techcrunch.com/2026/10/15/amazons-bee-launches-two-new-personal-ai-devices-starting-at-75/); [Bloomberg: Amazon debuts revamped Bee AI wearables with $12-a-month subscription](https://www.bloomberg.com/news/articles/2026-10-15/amazon-debuts-revamped-bee-ai-wearables-12-a-month-subscription-plan); [TechCrunch: Panos Panay sees a future where AI devices replace the need to use apps](https://techcrunch.com/2026/10/15/amazon-devices-chief-panos-panay-sees-a-future-where-ai-devices-replace-the-need-to-use-apps/)

### 2026-10-15: Anthropic makes Code Review for Claude Code generally available, with up to $50,000 in Enterprise credits

**; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-10-15-claude-code-review-general-availability/

On Oct 15, 2026 Anthropic moved Code Review, the automated multi-agent pull-request reviewer in Claude Code, out of research preview for Claude Team and Enterprise plans. Most reviews now take 6–10 minutes (about 20 in preview); eligible Enterprise organisations can claim $10,000–$50,000 in promotional credits until Jan 15, 2027.

- Plans: Claude Team and Enterprise; admins install a GitHub App, pick repositories and set Quick, Standard or Deep depth per repository; reviews run on new pull requests
- Speed: most reviews finish in 6–10 minutes, down from about 20 during the preview (Anthropic)
- Cost: reviews billed on token usage, about $15–25 each during the preview (TNW, docs)
- Credits: $10,000–$50,000 for eligible Enterprise organisations by active seats, valid 30 days after claim, claim window to Jan 15, 2027; not for organisations connected to GitHub Enterprise Cloud with data residency (.ghe.com)
- Anthropic's own figures (TNW): share of its PRs with substantive comments rose from 16% to 54%; engineers marked under 1% of comments incorrect. Docs warn 'a clean review does not mean the code is sound'

#### What happened
Anthropic announced general availability of Code Review for Claude Code on Oct 15, 2026. The reviewer runs several agents on
each pull request and posts comments in GitHub. Read here through The Next Web's report, which links the official post and the
credits support article; the official pages were not opened.

#### Why it matters
Automated review is becoming the bottleneck tool for AI-written code: two days earlier Anthropic described its own source-control
system built for tens of thousands of merges a day. Performance numbers are Anthropic's own.

Sources: [Anthropic: Code Review is now generally available](https://claude.com/resources/articles/code-review-is-now-generally-available); [Claude Code docs: Code Review](https://code.claude.com/docs/en/code-review); [Claude support: free Code Review promotional credits for Claude Enterprise](https://support.claude.com/en/articles/17603594-free-code-review-promotional-credits-for-claude-enterprise); [The Next Web: Anthropic makes Code Review generally available, with $50,000 in credits attached](https://thenextweb.com/news/anthropic-code-review-ga-eu-residency)

### 2026-10-15: OpenAI launches the Rosalind Academic Plan: sponsored GPT-Rosalind access for academic life-science labs

**; OpenAI; Products; confidence medium; https://postcutoff.com/e/2026-10-15-openai-rosalind-academic-plan/

On Oct 15, 2026 OpenAI introduced the Rosalind Academic Plan, "offering eligible academic labs sponsored access to GPT-Rosalind and tools to advance life sciences research" (OpenAI news feed). GPT-Rosalind, OpenAI's life-sciences research model released in April 2026, had until then been offered mainly through a trusted-access programme to companies and institutions.

- Announced Oct 15, 2026 (19:00 UTC) in OpenAI's post 'Bringing GPT-Rosalind to academic research labs'
- OpenAI's description: 'offering eligible academic labs sponsored access to GPT-Rosalind and tools to advance life sciences research'
- Not verified: eligibility rules, the amount of sponsorship and named labs (the post itself returned HTTP 403 here)
- OpenAI on X (Oct 15, ~302k views): the Rosalind Academic Plan 'helps scientific labs bring AI into the work of understanding biology, exploring ideas, and deciding what to test next', with sponsored access for eligible labs at U.S. universities

#### What happened
OpenAI announced a Rosalind Academic Plan giving eligible academic labs sponsored access to GPT-Rosalind and related tools. Only
the title and one-sentence description in OpenAI's RSS feed were read; the post was blocked (HTTP 403), so the details are not
yet recorded here.

#### Why it matters
GPT-Rosalind is OpenAI's specialised life-sciences model. Sponsored academic access widens it beyond pharma partners and
trusted-access users.

Sources: [OpenAI on X: Rosalind Academic Plan](https://x.com/OpenAI/status/2110841898076418336); [OpenAI: Bringing GPT-Rosalind to academic research labs](https://openai.com/index/introducing-the-rosalind-academic-plan); [OpenAI news RSS (description of the post)](https://openai.com/news/rss.xml)

### 2026-10-16: ChatGPT's GPT-6 Astra Pro gets faster responses, better plugin use and parallel agents for Pro users

**; OpenAI; Products; confidence high; https://postcutoff.com/e/2026-10-16-chatgpt-gpt-6-astra-pro-parallel-agents/

On Oct 16, 2026 OpenAI said GPT-6 Astra Pro in ChatGPT was upgraded with faster responses and better use of plugins and tools, and that "multiple agents can tackle different parts of a task in parallel", with each agent's progress visible and finished sub-tasks open for review. It is available to Pro subscribers in the Chat tab. The X post passed 1M views within a day.

- OpenAI on X (Oct 16, 20:27 UTC): 'GPT-6 Astra Pro in ChatGPT got an upgrade with faster responses and improved use of plugins and tools. And now, multiple agents can tackle different parts of a task in parallel. You can follow each agent's progress and open completed tasks to review their work. Now available to Pro users in the Chat tab.'
- Reach: about 1.0M views by Oct 17 (fxtwitter)
- Not stated: speed figures, which plugins, how many agents run at once, or whether the change reaches the API (GPT-6 Astra Pro on OpenRouter is the same model with reasoning mode 'pro', per our model file)

#### What happened
OpenAI announced the update only in a short X post; the ChatGPT release notes were not accessible to us. Parallel sub-agents inside a
single chat bring to ChatGPT's top tier the multi-agent pattern already common in coding agents.

Sources: [OpenAI on X: GPT-6 Astra Pro upgrade with parallel agents (Oct 16)](https://x.com/OpenAI/status/2111192259073094035)

### 2026-10-16: ChatGPT Slides rolls out in Space: collaborative AI-built decks for Pro, Business and Enterprise; GPT-6 Pro gets parallel agents

**; OpenAI; Products; confidence medium; https://postcutoff.com/e/2026-10-16-chatgpt-slides-space/

On Oct 16, 2026 OpenAI's ChatGPT release notes announced Slides, promised at DevDay (Sept 29): eligible Pro, Business and Enterprise users with Space access can have ChatGPT build a deck from a conversation, files, an outline or an example deck, edit it with teammates in real time, present it inside ChatGPT and export it to PowerPoint, PDF or Google Slides. The same notes reportedly gave GPT-6 Pro multiple agents that handle parts of a task in parallel. It extends OpenAI's push into Microsoft Office's and Google Workspace's territory.

- Release notes dated Oct 16, 2026 (help.openai.com; blocked for us, read via AI Brief and other summaries): rolling out gradually on web and desktop to eligible Pro, Business and Enterprise users with Space access; started by asking ChatGPT in Work to create a presentation in Space
- Features: start from a conversation, files, an outline or an example deck; edit text, images, shapes and charts directly or by asking ChatGPT; real-time co-editing with invited teammates and shared comments; slideshow inside ChatGPT
- Export: PowerPoint, PDF or Google Slides on the web; on desktop decks open in PowerPoint or Keynote if installed
- Not at launch: Plus (and apparently Free) accounts; decks have view and edit roles only (no comment-only role, unlike Pages), per Beri.net
- Same release-notes batch (per AI Brief): GPT-6 Pro gained 'multiple agents that can handle parts of a task in parallel'; no further detail
- Background: Slides, Pages and Space were announced at DevDay on Sept 29, 2026 ('in the coming weeks')

#### What happened
Two and a half weeks after DevDay, OpenAI began rolling out Slides, the presentation part of ChatGPT's Space workspace. ChatGPT builds a deck
from a chat or files; people and ChatGPT then edit it together, and it can be exported to the Office and Google formats. Our sources are
summaries of OpenAI's release notes; we could not open the notes or the feature page ourselves, hence confidence medium.

#### Why it matters
With Pages and Slides, ChatGPT now covers documents and presentations natively, competing with Microsoft 365 Copilot, Google Workspace
with Gemini and Anthropic's Claude docs and slides tools (Sept 2026).

Sources: [OpenAI: ChatGPT release notes (Oct 16, 2026 entry; not readable for us)](https://help.openai.com/en/articles/6825453-chatgpt-release-notes); [OpenAI: ChatGPT Slides feature page (403 for us)](https://chatgpt.com/features/slides/); [AI Brief: OpenAI adds Slides and parallel agents to ChatGPT (Oct 17)](https://dailyaibrief.com/news/openai-adds-collaborative-slide-creation-to-chatgpt-qNFmvJtV); [Beri.net: ChatGPT Slides puts your decks under Space's sharing rules](https://www.beri.net/article/chatgpt-slides-launch-pro-business-enterprise-space-sharing-powerpoint-google-slides-export-vs-copilot-gemini)

### 2026-10-16: DeepSeek launches standalone Search and Fetch APIs for agents, built on its own search engine ($0.75 per 1,000 calls)

**; DeepSeek; Products; confidence high; https://postcutoff.com/e/2026-10-16-deepseek-search-fetch-api/

DeepSeek added two raw JSON endpoints to its API, POST /v1/tools/search and POST /v1/tools/fetch, giving agents and RAG apps live web search (powered by DeepSeek's in-house search engine) and page fetching as Markdown. Each costs $0.75 per 1,000 successful requests (Pandaily reports 5 yuan per 1,000 calls), with 5 QPS per account; calls made through DeepSeek Harness, Claude Code or Codex are free until billing starts on Nov 1, 2026.

- Endpoints (base https://api.deepseek.com, bearer auth): POST /v1/tools/search and POST /v1/tools/fetch (DeepSeek API docs)
- Search: query of 1–128 characters with operators site:, -site:, "phrase", -term, lang:, after:, before:; count 1–50 (default 10); optional ori_question and IP-based localization; results carry rerank_score (0–3), authority (0–4), ~100–200-character snippet, ~600–800-character summary for models, optional full content
- Fetch: URL up to 8,192 characters; returns title and Markdown content; 403 if refused by security policy
- Price: $0.75 per 1,000 successful requests for each API; only HTTP 200 responses with results or content are billed; count does not change the price; Pandaily gives the CNY price as 5 yuan per 1,000 calls
- Free period: server-side tool calls via DeepSeek Harness, Claude Code or Codex are free for now; per-request billing starts 00:00 Beijing time, Nov 1, 2026
- Rate limits: 5 QPS each per account (Fetch also per domain), separate from model concurrency limits
- Launch date: the docs banner says 'now available' without a date; Pandaily reported it on Oct 17 (Beijing time); dated Oct 16 here

#### What happened
DeepSeek's API change log now opens with a banner announcing the Search / Fetch API. Unlike search tools bundled inside a chat model,
these are standalone endpoints any agent framework can call. The free tier for calls made from Claude Code and Codex suggests DeepSeek
wants its search to be the default web tool for third-party coding agents running DeepSeek models.

#### Why it matters
Web search had been an add-on bought from Western providers (Bing, Brave, Exa, Tavily) or from Chinese search giants. A frontier lab
selling its own search index at a low per-call price adds another piece of the agent stack under Chinese control, at a time when China is
proposing export controls on AI agents.

Sources: [DeepSeek API docs: Search / Fetch API guide](https://api-docs.deepseek.com/guides/search); [Pandaily: DeepSeek adds standalone Search and Fetch APIs for agents, priced at 5 yuan per 1,000 calls](https://pandaily.com/deepseek-api-search-fetch-endpoints-5-yuan-per-1000-requests-5-qps-agents-rag)

### 2026-10-18: Sam Altman's World launches a World ID suite for banks, including AgentKit for delegating 'proof of human' credentials to AI agents

**; World, Tools for Humanity; Products; confidence high; https://postcutoff.com/e/2026-10-18-world-id-suite-banks-agent-proofs/

On Oct 18, 2026 at Money20/20, World (Tools for Humanity, co-founded by Sam Altman) launched World ID for banks and financial firms, its first industry-specific offering. It bundles the World ID wallet, World ID for Zoom (now generally available, to show a caller is human rather than a deepfake) and World ID for Agents (AgentKit), which lets a person delegate their "Proofs" to an AI agent.

- Announced by co-founder Alex Blania at Money20/20 on Sunday Oct 18; components: World ID app (identity wallet for 'Proofs'), World ID for Zoom (GA after a beta since April), World ID for Agents / AgentKit (delegate Proofs to AI agents that present them to third parties)
- Scale: 18M+ people hold a Proof of Human; AgentKit launched in beta in March on Coinbase's x402 protocol
- Design partners: OpenExchange, Bain Capital, VanEck Funds, Outtake; no bank customer named and no pricing disclosed
- World's launch post opens with the 2024 Arup case, in which an employee transferred $25.6M after a video call where every other participant was a deepfake
- Pitch cites a Deloitte forecast that US AI-enabled fraud losses could reach $40B by 2027 (from ~$12.3B in 2023). Tools for Humanity president Sandro Herbig: 'an AI-safe trust layer across their internal and external systems'

#### What happened
World turned its proof-of-personhood system into a package for financial institutions, combining deepfake-resistant video calls with a
way for verified humans to hand credentials to AI agents acting for them.

#### Why it matters
As agents start transacting for people, banks need to know which agent acts for which verified human; World is positioning its iris-based
ID as that layer, though no bank has yet been named as a user.

Sources: [The Block: World launches World ID suite for banks](https://www.theblock.co/news/ecosystems/2026-10-18-world-launches-world-id-suite-banks-418869); [World Network on X: Introducing World ID for banking and finance (Oct 19 UTC; ~76k views)](https://x.com/worldnetwork/status/2111995130400178238); [World ID on X: 'World ID is having a moment' (Oct 19)](https://x.com/world_id/status/2112002312881557584)

### 2026-10-19: OpenAI says its latest models are stronger at building and migrating COBOL, Fortran and Visual Basic code

**; OpenAI; Products; confidence medium; https://postcutoff.com/e/2026-10-19-openai-low-resource-programming-languages/

On Oct 19, 2026 OpenAI posted that its "latest model advances" bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran and Visual Basic, aimed at banks, governments and businesses modernising legacy systems, and asked enterprise teams which languages and workflows to tackle next. It names no new model; the linked blog post could not be read here, so any benchmark figures are unverified.

- OpenAI on X (Oct 19, 22:30 UTC; ~222k views by Oct 20 01:30 CEST, fxtwitter): 'Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic.'
- Pitch: 'These languages power software that banks, governments, and businesses depend on. Progress here means understanding decades of business logic and helping developers modernize those systems while preserving what works.'
- Request: 'If your team is working on these systems, tell us which languages and workflows we should tackle next.'
- Blog post 'Advancing coding in low-resource programming languages' (openai.com/index/low-resource-programming-languages) was blocked for us (403 / JavaScript wall), so its models, benchmarks and numbers are not recorded here
- Same day OpenAI open-sourced 'MultiLangBench', 'An eval based on SQLite that benchmarks language models in multiple languages' (GitHub repo description); whether it belongs to this announcement is not confirmed
- Context: in February 2026 an Anthropic post on Claude modernising COBOL coincided with a sharp fall in IBM's stock (see related entry)

#### What happened
On the evening of Oct 19, 2026 (US time) OpenAI announced on X that its latest models are better at writing and migrating code
in low-resource languages such as COBOL, Fortran and Visual Basic, and linked a blog post. The X post names no model and gives
no numbers.

#### Why it matters
Legacy COBOL and Fortran systems run much of banking and government. Labs now compete for enterprise modernisation work, a
market whose stocks have reacted to such announcements before.

Not verified: the blog post could not be read (blocked), so which models, benchmarks and results it reports is unknown here.

Sources: [OpenAI on X: low-resource programming languages](https://x.com/OpenAI/status/2112310345863930168); [OpenAI: Advancing coding in low-resource programming languages (not read: blocked)](https://openai.com/index/low-resource-programming-languages); [GitHub: openai/MultiLangBench](https://github.com/openai/MultiLangBench)

### 2026-10-20: ElevenLabs launches Eleven Realtime v1 Max and Mini, speech-to-speech voice-agent models compatible with OpenAI's Realtime API

**; ElevenLabs; Products; confidence high; https://postcutoff.com/e/2026-10-20-eleven-realtime-speech-to-speech/

On Oct 20, 2026 ElevenLabs released two speech-to-speech models for voice agents: v1 Max for multi-step, tool-heavy work and v1 Mini for cheap conversation, both in 90+ languages with any of 17,000+ voices. The API mirrors OpenAI's Realtime API, so switching is mostly a change of endpoint and model name. Artificial Analysis measured Max at 98.8% on Big Bench Audio and 87.1% on tool use.

- Published Oct 20, 2026; two models: Eleven Realtime v1 Max (complex support and lead-qualification flows, heavy tool calling) and v1 Mini (assistants, tutors, characters)
- Artificial Analysis figures quoted by ElevenLabs: Big Bench Audio Max 98.8% / Mini 75.9%; Full-Duplex-Bench Mini 95.8% / Max 94.1%; Speech Agent Arena Max 998 / Mini 988; tool use Max 87.1% / Mini 60.3%; time to first audio ~0.57 s (Max), ~0.62 s (Mini)
- OpenAI Realtime API compatible over WebSocket (not WebRTC); OpenAI's SDK works unchanged; runs separately from the ElevenAgents platform
- AA speech-to-speech index (read 2026-10-25): v1 Max #6 of 21 (77.3)

#### What happened
ElevenLabs, until now mostly a text-to-speech company, shipped end-to-end speech-to-speech models for voice agents and made them a
drop-in replacement for OpenAI's Realtime API.

#### Why it matters
Voice agents are shifting from speech-to-text, LLM and text-to-speech pipelines to single speech models; ElevenLabs is competing
directly with OpenAI on that layer.

Sources: [ElevenLabs: Eleven Realtime (Oct 20)](https://elevenlabs.io/blog/realtime); [ElevenLabs docs: OpenAI Realtime compatibility](https://elevenlabs.io/docs/eleven-agents/guides/openai-realtime-compatibility)

### 2026-10-21: Anthropic's Claude Security now scans repositories continuously and tracks fixes; over one million vulnerabilities found since the April beta

***; Anthropic; Products; confidence high; https://postcutoff.com/e/2026-10-21-claude-security-continuous-scanning/

On Oct 21, 2026 Anthropic made Claude Security, its code-vulnerability scanner, continuous: after a baseline scan it re-scans repositories as code changes, tracks findings until a fix ships and reopens them if they return. It runs on Anthropic's full model family including Claude Mythos, without the Cyber Verification Program. Anthropic says thousands of organisations have used it since the April beta and found over one million vulnerabilities. Public beta for Claude Enterprise, with $25,000 a month in free credits for two months.

- Announced Oct 21, 2026 (claude.com article and @claudeai post, ~680k views on Oct 22); Claude Security itself entered public beta in April 2026
- Models: 'Anthropic's full model family, including Claude Mythos', with no Cyber Verification Program access required
- Continuous scanning: connect repositories via the Claude GitHub app (incl. GitHub Enterprise Server) or a GitLab/Git token; a baseline scan, then re-scans of changes chosen by risk and complexity; one project can span several repositories
- Fix tracking: open findings are re-checked against changed code and marked fixed when a fix ships (from Claude Code or elsewhere); they reopen automatically if detected again
- Triage: per-project knowledge base and editable threat model; each finding gets a severity judged against the threat model, an exploit-path explanation and optionally an attack-path diagram
- Integrations: API to list, read, resolve and reopen findings and manage projects; personal access tokens and webhooks; MCP connector for Claude.ai and Claude Code; role-based access controls
- Scale claim (Anthropic): thousands of organisations since April; 'over one million vulnerabilities' found
- Offer: $25,000 in free usage credits per month for two months, covering usage through Dec 31, 2026; public beta for Claude Enterprise customers, enabled by admins
- Customer quote (IonQ's Jon Donovan): 'Claude Security found many material issues that a traditional pen test conducted at the same time missed'
- Context: launched the same day OpenAI and Wiz disclosed PHP2Shell, found with GPT-5.6 Sol and Gemini 4 Argon

#### What happened
Anthropic turned Claude Security from a one-off scanner into a continuous one. Once a repository is connected, Claude keeps its
findings in step with the code: it rescans what changes, closes findings when they are fixed, and reopens them if they come back.
The product runs on Claude Mythos among other models; access to Mythos-level vulnerability finding normally requires Anthropic's Cyber
Verification Program, which this product does not.

The April 2026 beta launch is not yet an entry of its own; this entry rests on the Oct 21 article, which gives the April date.

#### Why it matters
Frontier labs are now selling automated vulnerability discovery to defenders at scale while the same capability finds internet-wide
bugs such as PHP2Shell. The "one million vulnerabilities" figure (Anthropic's own, unaudited) is a measure of how much latent risk
these tools surface, and of the patching load they create.

Sources: [Claude: Claude Security now scans continuously and tracks fixes (Oct 21, 2026)](https://claude.com/resources/articles/claude-security-now-scans-continuously-and-tracks-fixes); [Claude Security product page](https://claude.com/product/claude-security); [@claudeai on X: Claude Security now keeps up with your code as it changes](https://x.com/claudeai/status/2112962248218148878)

### 2026-10-22: Anthropic: Claude web, desktop and mobile apps will run in auto mode by default on Pro, Max and Team plans; Anthropic-managed Claude on Google Cloud goes GA

***; Anthropic, Google; Products; confidence high; https://postcutoff.com/e/2026-10-22-claude-apps-auto-mode-default/

On Oct 22, 2026 Anthropic announced that, starting the following week, sessions in the Claude web, desktop and mobile apps will run in auto mode by default on Pro, Max and Team plans: instead of asking users to approve each step, an automated check blocks risky actions, and users can switch back to manual. Auto mode had been the default in Claude Code since Aug 14. The same evening Anthropic and Google made "Claude Models on Google Cloud (Anthropic managed)" generally available.

- @claudeai (Oct 22, ~1.64M views): 'Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans. Rather than asking you to approve each step, an automated check blocks risky actions. Switch back to manual anytime.'
- Precedent quoted in the post: auto mode became the default permission mode in Claude Code for Pro, Max and Team users on Aug 14; a separate classifier reviews shell commands and actions, and in Anthropic's testing it 'caught 89% of dangerous commands. Manual approval caught 14%'
- Google Cloud (Oct 23 blog): 'Claude Models on Google Cloud (Anthropic managed)' is GA: Anthropic operates the stack as data processor under its Commercial Terms on Google Cloud infrastructure, with Google Cloud IAM, Cloud Audit Logging, Google Cloud billing and commitment drawdown, and day-zero access to Claude Platform features (Managed Agents beta, skills, code execution); models listed: Claude Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5. The Google-managed option keeps VPC Service Controls, CMEK and data residency
- Reach: the Google Cloud GA post on @claudeai had ~468k views

#### What happened
Anthropic is moving its consumer and team apps, where Claude increasingly runs agentic sessions, to the same classifier-gated permission
model it made the default in Claude Code in August. Users stop approving each step; a separate model blocks actions it judges risky.

#### Why it matters
Default settings decide what most people actually run. The change hands routine oversight of agent actions in Claude's apps from people
to an automated check, a trade-off Anthropic justifies with its own figures (the classifier caught far more dangerous commands than
fatigued humans did). It comes in a week when rogue-agent behaviour (banks, OpenAI's monitoring) dominated the news.

Sources: [Claude on X: auto mode by default in Claude apps (Oct 22, 2026)](https://x.com/claudeai/status/2113360845409583307); [Claude on X: Claude Models on Google Cloud (Anthropic managed) generally available](https://x.com/claudeai/status/2113380122468454765); [Google Cloud: Expanding Claude availability and deployment choice on Google Cloud](https://cloud.google.com/blog/products/ai-machine-learning/expanding-claude-availability-on-google-cloud)

## Models

### Sesame CSM-1B (Conversational Speech Model) (Sesame, 2025-03-13, current)

Type audio/speech, open weights. https://postcutoff.com/m/sesame-csm-1b/
Access: Hugging Face `sesame/csm-1b` https://huggingface.co/sesame/csm-1b; Transformers `sesame/csm-1b`; Sesame app (Maya, Miles, Simone, Charlie — hosted larger models) https://www.sesame.com/
- Context-conditioned conversational TTS: Llama backbone + audio decoder emitting Mimi audio codes; generates speech conditioned on prior conversation audio/text so prosody fits the dialogue; voice prompting via context segments.

### Speechmatics Linden 1 (Agent STT) (Speechmatics, 2026-09-17, current)

Type audio/speech. https://postcutoff.com/m/speechmatics-linden-1/
Access: Speechmatics Agent STT API `linden-1`; Pipecat https://www.speechmatics.com/voice-agents; LiveKit https://docs.livekit.io/agents/models/stt/speechmatics/
- STT output shaped for LLM voice agents: Returns speaker-attributed segments with turn messages instead of a running word stream; finalizes segments in under 350 ms; 55+ languages; custom vocabulary up to 1,000 terms; live diarization and speaker ID.
- Low semantic error on Pipecat benchmark: 1.05% pooled semantic error rate and 369 ms median finalization on the Pipecat STT benchmark (23 streaming models), on the speed/accuracy Pareto frontier, per Speechmatics.

### Gemini 3.8 Flash-Lite TTS (Google DeepMind, 2026-09-22, current)

Type audio/speech. https://postcutoff.com/m/gemini-3-8-flash-lite-tts/
Access: Gemini API `gemini-3.8-flash-lite-tts`; Google AI Studio https://aistudio.google.com
- Low-cost everyday TTS in 101 languages: Built to replace gemini-3.1-flash-tts-preview for everyday single-speaker generation; 101 languages; audio out at $6 per 1M tokens vs $9 for Gemini 3.8 Flash TTS (intro prices).
- Voice design and stored voice replication: Custom Voice design personas, prebuilt voices plus the Extended Voice Library, and persistent stored replicated voices (optional stateless keys).
- Hume VoiceEQ 0.914: Scored 0.914 on Hume's blind Real-World VoiceEQ leaderboard, just behind Gemini 3.8 Flash TTS (0.920) and ahead of Cartesia Sonic 3.6 (0.840).

### Eleven Realtime v1 Max / Mini (ElevenLabs, 2026-10-20, current)

Type audio/speech. https://postcutoff.com/m/eleven-realtime-v1-max/
Access: ElevenLabs Realtime API (WebSocket, OpenAI Realtime-compatible) `eleven_realtime_v1_max`; ElevenLabs Realtime API `eleven_realtime_v1_mini`; ElevenAgents https://elevenlabs.io/realtime
- Reasoning speech-to-speech for tool-heavy agents (Max): Max reasons before responding; 87.1% task success on tool-use tasks (Mini 60.3%) and 98.8% on Big Bench Audio (Mini 75.9%), per ElevenLabs citing Artificial Analysis' independent evaluation.
- Drop-in OpenAI Realtime API compatibility: The Realtime API follows OpenAI's Realtime spec, so the OpenAI SDK works by changing endpoint and model name; WebSocket only at launch (no WebRTC).
- 90+ languages and 17,000+ library voices: Realtime conversation in over 90 languages with any of 17,000+ ElevenLabs voices; time to first audio ~0.57 s (Max) and ~0.62 s (Mini).
- #6 on Artificial Analysis Speech-to-Speech Index: Max scored 77.3 on the AA Speech-to-Speech Index (#6 of 21), Arena Elo 997, time to first audio 0.57 s; Mini 64.4 (#17), read 2026-10-25.

## Videos

- 2026-10-16, ABC News: [Elon Musk unveils Tesla's new Roadster, and it can fly](https://www.youtube.com/watch?v=BYWy6SyJRZo) https://postcutoff.com/v/abc-news-tesla-roadster-unveiled-can-fly/
- 2026-10-16, carwow: [New Tesla Roadster is here, and it's INSANE!](https://www.youtube.com/watch?v=4bbfRQDC6cs) https://postcutoff.com/v/carwow-new-tesla-roadster-is-here-insane/
- 2026-10-16, DragTimes: [INSIDE LOOK! I saw the Flying Tesla Roadster in Person and it was CRAZY](https://www.youtube.com/watch?v=_lq-DpCE9rU) https://postcutoff.com/v/dragtimes-saw-flying-tesla-roadster-in-person/
- 2026-10-16, Everyday Chris: [I Went to Tesla's Roadster Event… here's what they didn't show you](https://www.youtube.com/watch?v=pomFE55c_Sw) https://postcutoff.com/v/everyday-chris-tesla-roadster-event-what-they-didnt-show/
- 2026-10-16, Hagerty: [The FLYING Tesla Roadster explained by the man who engineered it](https://www.youtube.com/watch?v=22rvpUecLts) https://postcutoff.com/v/hagerty-flying-tesla-roadster-lars-moravy-explains/
- 2026-10-16, Marques Brownlee: [The NEW Tesla Roadster: My Honest Reaction](https://www.youtube.com/watch?v=scwlR0QHGOk) https://postcutoff.com/v/mkbhd-new-tesla-roadster-honest-reaction/
- 2026-10-16, MotorTrend Channel: [Full Walkaround with Chief Designer of Tesla Roadster](https://www.youtube.com/watch?v=y7oKy09ALfM) https://postcutoff.com/v/motortrend-tesla-roadster-walkaround-chief-designer/
- 2026-10-16, NBC News: [Tesla unveils its next generation 'flying' Roadster](https://www.youtube.com/watch?v=4wCGk5VaGmQ) https://postcutoff.com/v/nbc-news-tesla-unveils-next-generation-flying-roadster/
- 2026-10-16, REC Anything: [All New 2028 Tesla Roadster can Fly - World Premiere and First Look](https://www.youtube.com/watch?v=d2jH2x1HWxE) https://postcutoff.com/v/rec-anything-2028-tesla-roadster-can-fly-first-look/
- 2026-10-16, SlashGear: [Was The Tesla Roadster Flip Real, Or An AI-Generated Stunt?](https://www.youtube.com/watch?v=6phDTHxrei8) https://postcutoff.com/v/slashgear-was-tesla-roadster-flip-real-or-ai/
- 2026-10-16, TaouyRev: [The Viral Tesla Flying Car Is Fake — But Tesla Really Did Show One](https://www.youtube.com/watch?v=o7ltL0f1IPY) https://postcutoff.com/v/taouyrev-viral-tesla-flying-car-fake-but-tesla-showed-one/
- 2026-10-16, Top Gear: [New Tesla Roadster: SpaceX Thrusters, 0.6sec 0-60mph & 330mph! | 4K](https://www.youtube.com/watch?v=JdkIOhFmbKI) https://postcutoff.com/v/top-gear-new-tesla-roadster-spacex-thrusters/
- 2026-10-16, YOUCAR: [NEW TESLA ROADSTER - It Can Fly!!! (SpaceX Technology)](https://www.youtube.com/watch?v=Fdhwkap2DFU) https://postcutoff.com/v/youcar-new-tesla-roadster-it-can-fly/
- 2026-10-15, CNET: [Tesla's Flying Roadster Revealed: Everything That Happened in 10 Minutes](https://www.youtube.com/watch?v=TWHXrLzUKJc) https://postcutoff.com/v/cnet-tesla-flying-roadster-revealed-10-minutes/
- 2026-10-15, news.com.au: [Tesla's new Roadster can FLY](https://www.youtube.com/watch?v=aN6strasoHA) https://postcutoff.com/v/news-com-au-teslas-new-roadster-can-fly/
- 2026-10-15, OpenAI: [Build more personal sites with Codex](https://www.youtube.com/watch?v=KJ_7KPKlO_Q) https://postcutoff.com/v/openai-build-more-personal-sites-with-codex/
- 2026-10-15, Tesla: [Roadster Event](https://www.youtube.com/watch?v=qt5p-_tE8d0) https://postcutoff.com/v/tesla-roadster-event-livestream-2026-10-15/
- 2026-10-15, Tesla: [Roadster | The Last Best Driver's Car](https://www.youtube.com/watch?v=cvKIR-c1SsU) https://postcutoff.com/v/tesla-roadster-last-best-drivers-car/
- 2026-10-15, Tesla: [The Kickflip | Roadster](https://www.youtube.com/watch?v=r5vsHHPVhJg) https://postcutoff.com/v/tesla-roadster-the-kickflip/
- 2026-10-15, Tesla: [The Ramp | Roadster](https://www.youtube.com/watch?v=7p5LxFdc-Qg) https://postcutoff.com/v/tesla-roadster-the-ramp-live-demo/
- 2026-10-15, VideoFromSpace (Space.com): [Tesla Unveils 'Flying' Roadster with SpaceX Thrusters](https://www.youtube.com/watch?v=eqOfmtTJD1A) https://postcutoff.com/v/videofromspace-tesla-flying-roadster-spacex-thrusters/
- 2026-10-14, kaye.creatives: [How To Make Insane Motion Graphics With Opus 5.5 (Claude Motion)](https://www.youtube.com/watch?v=AsvN_feasyY) https://postcutoff.com/v/kaye-creatives-claude-motion-motion-graphics/
- 2026-10-14, Tristen O'Brien: [Claude Just Dropped MOTION. (Get Ahead of 95% in 11 Minutes.)](https://www.youtube.com/watch?v=yf2OwchM9nw) https://postcutoff.com/v/tristen-obrien-claude-just-dropped-motion/
- 2026-10-13, Apple: [The new Apple HomeView: Keep the whole home connected, entertained, and in the know.](https://www.youtube.com/watch?v=QeQAXuQpGEA) https://postcutoff.com/v/apple-homeview-keep-whole-home-connected/
- 2026-10-13, Apple: [Welcome Apple HomeView: Designed for life at home.](https://www.youtube.com/watch?v=hkrYiZ_g9tw) https://postcutoff.com/v/apple-welcome-apple-homeview/
- 2026-10-13, Higgsfield AI: [How I Make Motion Graphics Inside Claude Motion (Opus 5.5)](https://www.youtube.com/watch?v=5HYYXYYCrL4) https://postcutoff.com/v/higgsfield-motion-graphics-inside-claude-motion-opus-5-5/
- 2026-10-13, iJustine: [Apple's new HomeView and accessories!](https://www.youtube.com/watch?v=L-cnJ2ERh_Y) https://postcutoff.com/v/ijustine-apple-homeview-and-accessories/
- 2026-10-13, LMGClips: [Google's New Laptop Almost Ended the WAN Show](https://www.youtube.com/watch?v=ahXB3n5uAKA) https://postcutoff.com/v/lmgclips-googlebook-almost-ended-wan-show/
- 2026-10-13, Linus Tech Tips: [Everything Apple announced in under 10 mins - Welcome Home 2026](https://www.youtube.com/watch?v=TD74a5g5rp0) https://postcutoff.com/v/ltt-everything-apple-announced-welcome-home-2026/
- 2026-10-13, MacRumors: [Everything Apple Announced at Its Welcome Home Event (Apple HomeView, and More)](https://www.youtube.com/watch?v=yByyCBngrGg) https://postcutoff.com/v/macrumors-everything-apple-announced-welcome-home/
- 2026-10-13, Marques Brownlee: [Apple Homeview Impressions: Not Convinced!](https://www.youtube.com/watch?v=QVv3mQzlrN0) https://postcutoff.com/v/mkbhd-apple-homeview-impressions/
- 2026-10-13, OpenAI: [Work with your team in ChatGPT Space](https://www.youtube.com/watch?v=G-9Y2xNiDKQ) https://postcutoff.com/v/openai-work-with-your-team-in-chatgpt-space/
- 2026-10-13, The Verge: [HomeView: Apple just got serious about the smart home](https://www.youtube.com/watch?v=bHQ-kdep7Og) https://postcutoff.com/v/the-verge-homeview-apple-serious-smart-home/
- 2026-10-12, OpenAI: [Code review is no longer the bottleneck](https://www.youtube.com/watch?v=IZXv6a78NDw) https://postcutoff.com/v/openai-codex-code-review-no-longer-the-bottleneck/
- 2026-10-10, Riley Brown: [Claude's New Updates Are Insane… But Grok Bot Is EVEN Bigger](https://www.youtube.com/watch?v=UFa8pXW9A3k) https://postcutoff.com/v/riley-brown-claude-updates-grok-bot-bigger/
- 2026-10-09, Android Authority: [Googlebook first impressions: Disaster](https://www.youtube.com/watch?v=VaxjK_zoZdA) https://postcutoff.com/v/android-authority-googlebook-first-impressions-disaster/
- 2026-10-09, Brock Mesarich | AI for Non Techies: [Anthropic Just Dropped Claude Motion, Dashboards + More (AI News)](https://www.youtube.com/watch?v=LlwhNla5eQg) https://postcutoff.com/v/brock-mesarich-claude-motion-dashboards-ai-news/
- 2026-10-09, Matt Wolfe: [AI News: Haiku-5.5, Visual ChatGPT, Grok's New Trick, and more!](https://www.youtube.com/watch?v=NUizyuGj-kE) https://postcutoff.com/v/matt-wolfe-ai-news-haiku-5-5-visual-chatgpt-grok/
- 2026-10-09, Paul J Lipsky: [Weekly AI Recap: GrokBot Just Did the Unthinkable, ChatGPT Looks Completely Different & More!](https://www.youtube.com/watch?v=S-IQRTRsJjo) https://postcutoff.com/v/paul-lipsky-weekly-ai-recap-grokbot-chatgpt-different/
- 2026-10-08, Claude: [Introducing Claude Dashboards and Claude Motion](https://www.youtube.com/watch?v=en0GuyhieQk) https://postcutoff.com/v/claude-introducing-dashboards-and-motion/
- 2026-10-08, Marques Brownlee: [WTF is a Googlebook?](https://www.youtube.com/watch?v=_0G4OG39XjI) https://postcutoff.com/v/mkbhd-wtf-is-a-googlebook/
- 2026-10-08, RemakeBench: [I Tested Opus 5.5 vs GPT-6 Astra, Sonnet 5.5, GPT-6.1 Sol](https://www.youtube.com/watch?v=KW6NBXdQLDQ) https://postcutoff.com/v/remakebench-opus-5-5-vs-gpt-6-astra-sonnet-sol/
- 2026-10-08, Snazzy Labs: [The New Siri Sucks. But It Shouldn't.](https://www.youtube.com/watch?v=sD5qgPb2PR0) https://postcutoff.com/v/snazzy-labs-new-siri-sucks-but-shouldnt/
- 2026-10-07, Christian Peverelli: [ChatGPT Just Changed FOREVER - Intelligent UI Explained](https://www.youtube.com/watch?v=pt5eC8gLbWI) https://postcutoff.com/v/christian-peverelli-chatgpt-intelligent-ui-explained/
- 2026-10-07, Claude: [What's new in Claude: September recap](https://www.youtube.com/watch?v=ICGS7LlRfpI) https://postcutoff.com/v/claude-whats-new-september-recap/
- 2026-10-07, Google Labs: [Introducing Playground: If you can think it, you can play it.](https://www.youtube.com/watch?v=Dxvrvfyicp0) https://postcutoff.com/v/google-labs-introducing-playground/
- 2026-10-07, OpenAI: [Build Plugins for ChatGPT](https://www.youtube.com/watch?v=LiNhqbetrbU) https://postcutoff.com/v/openai-build-plugins-for-chatgpt/
- 2026-10-07, OpenAI: [Introducing GPT-6 in ChatGPT with Intelligent UI](https://www.youtube.com/watch?v=IsL4dVezs18) https://postcutoff.com/v/openai-introducing-gpt-6-chatgpt-intelligent-ui/
- 2026-10-06, BBC News: [What's going on with Meta's smart glasses? | The Global Story](https://www.youtube.com/watch?v=ixYKDmNuenk) https://postcutoff.com/v/bbc-global-story-meta-smart-glasses/
- 2026-10-06, Claude: [Claude for Google Workspace™](https://www.youtube.com/watch?v=idJQMJwHtyM) https://postcutoff.com/v/claude-for-google-workspace/
- 2026-10-06, OpenAI: [Introducing the Decisions API](https://www.youtube.com/watch?v=FB6oCmrIj-Y) https://postcutoff.com/v/openai-introducing-the-decisions-api/
- 2026-10-05, Claude: [Mods in Claude Code: change how Claude Code works](https://www.youtube.com/watch?v=yUuFvL1lK4k) https://postcutoff.com/v/claude-mods-in-claude-code/
- 2026-10-05, Dave2D: [The Laptop for 71% of People](https://www.youtube.com/watch?v=ZedAs9P7yE8) https://postcutoff.com/v/dave2d-the-laptop-for-71-percent-of-people/
- 2026-10-05, OpenAI: [Meet the New Codex CLI](https://www.youtube.com/watch?v=PUBc2G0Tj5E) https://postcutoff.com/v/openai-meet-the-new-codex-cli/
- 2026-10-02, BBC News: [Meta's Muse: Cute AI Agent Or 'AI Trojan Horse'? | BBC News](https://www.youtube.com/watch?v=9VTuKGI54vQ) https://postcutoff.com/v/bbc-news-meta-muse-cute-ai-agent-or-ai-trojan-horse/
- 2026-10-02, Claude: [Build an App With Claude Design](https://www.youtube.com/watch?v=COJAZQM1aeQ) https://postcutoff.com/v/claude-build-an-app-with-claude-design/
- 2026-10-02, The PrimeTime: [A total disaster](https://www.youtube.com/watch?v=p0fybvFyOlM) https://postcutoff.com/v/primetime-a-total-disaster-openai-devday/
- 2026-10-01, Claude: [How an Anthropic designer uses Claude Slides](https://www.youtube.com/watch?v=bVXiNb9RyyI) https://postcutoff.com/v/claude-anthropic-designer-uses-claude-slides/
- 2026-10-01, Fireship: [The one OpenAI announcement that can actually make you money...](https://www.youtube.com/watch?v=No-JPdFvYWU) https://postcutoff.com/v/fireship-the-one-openai-announcement-that-can-actually-make-you-money/
- 2026-09-30, Silicon Valley Girl: [OpenAI Co-Founder: Start Building With AI Before You Feel Ready | Greg Brockman](https://www.youtube.com/watch?v=qy8Gr27yLMk) https://postcutoff.com/v/silicon-valley-girl-openai-co-founder-start-building-with-ai-before-you-feel-rea/
- 2026-09-29, Bloomberg Tech: [OpenAI Will Keep Driving Down Prices, Sam Altman Says](https://www.youtube.com/watch?v=Pj1DRuWVdSI) https://postcutoff.com/v/bloomberg-tech-openai-will-keep-driving-down-prices-sam-altman-says/
- 2026-09-29, BridgeMind: [OpenAI DevDay 2026](https://www.youtube.com/watch?v=IuV1gMP0-_g) https://postcutoff.com/v/bridgemind-openai-devday-2026/
- 2026-09-29, CNET: [OpenAI Dev Day 2026: Everything Announced in 15 Minutes](https://www.youtube.com/watch?v=GjN3xLDuc8o) https://postcutoff.com/v/cnet-openai-dev-day-2026-everything-announced-in-15-minutes/
- 2026-09-29, The Tonight Show Starring Jimmy Fallon: [Trump Launches AI-Powered Government Website, Karoline Leavitt Teases White House Tell-All Book](https://www.youtube.com/watch?v=3ApSBU8d3B4) https://postcutoff.com/v/fallon-trump-ai-government-website/
- 2026-09-29, Matthew Berman: [OpenAI COOKED](https://www.youtube.com/watch?v=Xc6ERvZM1NY) https://postcutoff.com/v/matthew-berman-openai-cooked/
- 2026-09-29, OpenAI: [Build faster with Ultrafast](https://www.youtube.com/watch?v=sKltfHvsDQM) https://postcutoff.com/v/openai-build-faster-with-ultrafast/
- 2026-09-29, OpenAI: [Meet the builders of our time - the Codex Originals.](https://www.youtube.com/watch?v=m8VkCbkFAKA) https://postcutoff.com/v/openai-codex-originals-builders/
- 2026-09-29, OpenAI: [Live from OpenAI DevDay 2026: Keynote](https://www.youtube.com/watch?v=Fls_onRviPM) https://postcutoff.com/v/openai-devday-2026-keynote/
- 2026-09-29, OpenAI: [Introducing dots, always-on agents built to handle everything.](https://www.youtube.com/watch?v=uXspbC2srEQ) https://postcutoff.com/v/openai-introducing-dots-always-on-agents-built-to-handle-everything/
- 2026-09-29, OpenAI: [Meet the all new Codex Cloud](https://www.youtube.com/watch?v=7Bv68f5szSU) https://postcutoff.com/v/openai-meet-the-all-new-codex-cloud/
- 2026-09-28, Kevin Stratvert: [Microsoft Copilot Just Got a Massive Upgrade](https://www.youtube.com/watch?v=KJ8Y2Nb9y9o) https://postcutoff.com/v/kevin-stratvert-new-microsoft-copilot-upgrade/
- 2026-09-27, ThrillSeeker: [Meta VR Glasses: First Impressions](https://www.youtube.com/watch?v=RYIXcxlGNn0) https://postcutoff.com/v/thrillseeker-meta-vr-glasses-first-impressions/
- 2026-09-25, Deirdre Bosa (with Yahoo Finance): [Microsoft's New Copilot: Why the AI Race Is Moving From Models to Products with CEO Satya Nadella](https://www.youtube.com/watch?v=eo4DF16uIIg) https://postcutoff.com/v/deirdre-bosa-nadella-new-copilot-models-to-products/
- 2026-09-25, Fireship: [Meta is pivoting again... everything you missed from Connect 2026](https://www.youtube.com/watch?v=c1rPlzxSZ8E) https://postcutoff.com/v/fireship-meta-connect-2026-pivoting-again/
- 2026-09-25, Microsoft Copilot: [The new Copilot: where AI-powered work comes together](https://www.youtube.com/watch?v=OgInADh5Tcs) https://postcutoff.com/v/microsoft-new-copilot-2026/
- 2026-09-25, Sources Podcast (Alex Heath): [Satya Nadella on the AI backlash, and the rise of agents](https://www.youtube.com/watch?v=giE1O1-e2DY) https://postcutoff.com/v/sources-nadella-ai-backlash-rise-of-agents/
- 2026-09-24, AI+ （エーアイプラス） (@AIPlus_AISchool): [Honnō-ji Incident explained in 60 seconds: video by Opus 5.5, voice by Gemini 3.8 TTS](https://x.com/AIPlus_AISchool/status/2102960034431070563) https://postcutoff.com/v/aiplus-opus-5-5-honnoji-incident-explainer-x/
- 2026-09-24, Bloomberg Podcasts: [Meta Introduces Palm-Sized Device to Use Muse AI on the Go](https://www.youtube.com/watch?v=6kkGXlwsMn0) https://postcutoff.com/v/bloomberg-meta-palm-sized-muse-device/
- 2026-09-24, Firstpost: [Meta's Charm: A Really Small AI Device That Wants to Replace Your Phone | FP Explains](https://www.youtube.com/watch?v=xSg4dnJUaCA) https://postcutoff.com/v/firstpost-meta-charm-small-ai-device/
- 2026-09-24, Adam Savage’s Tested: [Hands-On with Meta VR Glasses!](https://www.youtube.com/watch?v=Bzvdj8vhVlM) https://postcutoff.com/v/tested-hands-on-meta-vr-glasses/
- 2026-09-24, Tyriel Wood - VR Tech: [Meta VR Glasses - Full On Stage Live Demonstration](https://www.youtube.com/watch?v=S0rI6qABYMg) https://postcutoff.com/v/tyriel-wood-meta-vr-glasses-on-stage-demo/
- 2026-09-23, CNBC Television: [First look at Meta's new VR glasses](https://www.youtube.com/watch?v=l3nJcwuthDk) https://postcutoff.com/v/cnbc-first-look-meta-vr-glasses/
- 2026-09-23, CNET: [Hands-On With Meta VR Glasses and All of Meta's Other Glasses and AI](https://www.youtube.com/watch?v=wzUU-Jp3zoU) https://postcutoff.com/v/cnet-hands-on-meta-vr-glasses-and-ai/
- 2026-09-23, CNET: [New VR Glasses and More! Everything Announced at Meta Connect in Less Than 10 Minutes](https://www.youtube.com/watch?v=mFrDMb42zAs) https://postcutoff.com/v/cnet-meta-connect-2026-everything-in-10-minutes/
- 2026-09-23, Google DeepMind: [Create your own voices with Gemini 3.8 text-to-speech](https://www.youtube.com/watch?v=FL6mI_Br-mc) https://postcutoff.com/v/deepmind-gemini-3-8-tts-custom-voices/
- 2026-09-23, Joanna Stern: [Mark Zuckerberg on Muse, Audio-Only Glasses and Killer AI](https://www.youtube.com/watch?v=2cg56uF4hlc) https://postcutoff.com/v/joanna-stern-zuckerberg-muse-audio-glasses-killer-ai/
- 2026-09-23, Meta Developers: [Meta Connect 2026: Opening Keynote](https://www.youtube.com/watch?v=dnT9cVv3Spw) https://postcutoff.com/v/meta-connect-2026-developer-keynote/
- 2026-09-23, Meta: [Meta Connect Keynote 2026](https://www.youtube.com/watch?v=SdKFDIAGF24) https://postcutoff.com/v/meta-connect-2026-keynote/
- 2026-09-23, Meta VR: [Introducing Meta VR Glasses | Spring 2027](https://www.youtube.com/watch?v=xyD_bmMy5nQ) https://postcutoff.com/v/meta-introducing-meta-vr-glasses/
- 2026-09-23, VRelity: [Meta Connect 2026: Everything Revealed in 16 Minutes](https://www.youtube.com/watch?v=JvyVwlP_nlw) https://postcutoff.com/v/vrelity-meta-connect-2026-everything-revealed/
- 2026-09-17, Claude: [Projects are now a conversation with Claude](https://www.youtube.com/watch?v=5qt_aGyAsKk) https://postcutoff.com/v/claude-projects-conversation/
- 2026-09-16, Claude: [Claude Cowork and chat are now one Claude](https://www.youtube.com/watch?v=qMUf-jwSpMo) https://postcutoff.com/v/claude-cowork-and-chat-one-claude/
- 2026-09-16, Claude: [Meet Claude Slides, Claude Design and Claude Docs](https://www.youtube.com/watch?v=To5nrYqvR44) https://postcutoff.com/v/claude-meet-slides-design-docs/
- 2026-09-09, Apple: [Apple Event September ’26: Recapping announcements of iPhone Duo, iPhone 18 Pro, and more](https://www.youtube.com/watch?v=3fAHjTPvF1E) https://postcutoff.com/v/apple-event-september-2026-recap/
- 2026-09-09, Apple: [Apple Event September 9 2026: Introducing iPhone Duo and more](https://www.youtube.com/watch?v=39BalPDuTo0) https://postcutoff.com/v/apple-event-september-2026/
- 2026-01-12, Anthropic: [Introducing Cowork: Claude Code for the rest of your work](https://www.youtube.com/watch?v=UAmKyyZ-b9E) https://postcutoff.com/v/anthropic-introducing-cowork/

## Posts

- 2026-10-22, Claude: [](https://x.com/claudeai/status/2113360845409583307)
- 2026-10-22, Claude: [](https://x.com/claudeai/status/2113380122468454765)
- 2026-10-22, OpenAI Developers: [](https://x.com/OpenAIDevs/status/2113346642787086538)
- 2026-10-21, Claude: [Claude: Claude Security now keeps up with your code as it changes](https://x.com/claudeai/status/2112962248218148878)
- 2026-10-19, OpenAI: [](https://x.com/OpenAI/status/2112310345863930168)
- 2026-10-19, World ID: [](https://x.com/world_id/status/2112002312881557584)
- 2026-10-19, World: [](https://x.com/worldnetwork/status/2111995130400178238)
- 2026-10-16, Collin Rugg: [Collin Rugg: clip of Tesla's live Roadster ramp stunt with Musk's Peter Thiel 'flying cars' line](https://x.com/CollinRugg/status/2110901393238474863)
- 2026-10-16, Tesla: [Tesla: 'Do a kickflip!' — edited video of the Roadster's thruster roll over parked Teslas](https://x.com/Tesla/status/2110926164306931860)
- 2026-10-16, OpenAI: [](https://x.com/OpenAI/status/2111192259073094035)
- 2026-10-16, unusual_whales: [unusual_whales: 'The new Tesla Roadster flew over other Tesla vehicles' (7-second clip)](https://x.com/unusual_whales/status/2110910422358561071)
- 2026-10-15, OpenAI: [](https://x.com/OpenAI/status/2110841898076418336)
- 2026-10-15, OpenAI Developers: [](https://x.com/OpenAIDevs/status/2110832979656524283)
- 2026-10-15, Andy Jassy: [Andy Jassy announces Bee Frontier and Bee Eclipse](https://x.com/ajassy/status/2110729551571456391)
- 2026-10-07, OpenAI: [](https://x.com/OpenAI/status/2107895006350791071)
- 2026-10-04, Tibo (Thibault Sottiaux): [Sottiaux: every day for 28 days, a clear Codex/Work improvement or a full reset](https://x.com/thsottiaux/status/2106845241357824205)
- 2026-10-03, 9to5Google: [](https://x.com/9to5google/status/2106488912721957074)
- 2026-10-03, Gemini Updates ( fan ): [](https://x.com/geminiupda0kkb/status/2106324275829281101)
- 2026-10-02, Alexandr Wang: [](https://x.com/alexandr_wang/status/2106113742266089526)
- 2026-10-02, Alexandr Wang: [](https://x.com/alexandr_wang/status/2106116686654931161)
- 2026-10-02, Muse: [](https://x.com/muse/status/2106119592229884289)
- 2026-10-02, Federico Viticci: [](https://x.com/viticci/status/2106139174604816765)
- 2026-10-02, Nat Friedman: [Nat Friedman announces Muse Gadgets (open-source ESP32 firmware and Linux SDK for Muse hardware)](https://x.com/natfriedman/status/2106099383037309211)
- 2026-10-02, Nat Friedman: [Nat Friedman: Meta's Muse Home Link, an open-source USB-C dongle connecting Muse to home devices](https://x.com/natfriedman/status/2106099384891158562)
- 2026-10-01, Claude Devs: [Anthropic: You can now mod Claude Code](https://x.com/ClaudeDevs/status/2105721434807083061)
- 2026-09-29, OpenAI: [OpenAI: 'Introducing dots, powered by GPT-6 Astra.'](https://x.com/OpenAI/status/2104984504133918973)
- 2026-09-29, OpenAI: [OpenAI: Codex Security Cloud upgrade with Daybreak Blue included](https://x.com/OpenAI/status/2104987422308335828)
- 2026-09-29, Tibo: [](https://x.com/thsottiaux/status/2104987594719461796)
- 2026-09-29, OpenAI: [OpenAI: '10am PT, on the dot.' keynote teaser](https://x.com/OpenAI/status/2104934336206430527)
- 2026-09-28, OpenAI: [OpenAI teases DevDay: 'Get ready.' (20+ launches video)](https://x.com/OpenAI/status/2104651136699609518)
- 2026-09-28, Sam Altman: [Altman teases DevDay: 'found a new thing'](https://x.com/sama/status/2104661956879913457)
- 2026-09-25, Alexandr Wang: [Why We're Building Muse (X Article)](https://x.com/alexandr_wang/status/2103551714536439951)
- 2026-09-25, Jacob Andreou: [The New Copilot (X Article by Jacob Andreou)](https://x.com/jacobandreou/status/2103470116906176622)
- 2026-09-25, Omar Shahine: [Omar Shahine: Microsoft Autopilot (formerly Scout) is built on OpenClaw](https://x.com/OmarShahine/status/2103480227561079264)
- 2026-09-25, Peter Steinberger 🦞: [Steinberger: Microsoft shipped the new Copilot on top of OpenClaw](https://x.com/steipete/status/2103491173927272531)
- 2026-09-25, Deirdre Bosa: [Deirdre Bosa (CNBC): the new Copilot is Microsoft's bet that the race moves from models to products](https://x.com/dee_bosa/status/2103454471061688347)
- 2026-09-25, tae kim: [Nadella wants Autopilot for consumers too (per The Verge interview)](https://x.com/firstadopter/status/2103508672106188846)
- 2026-09-25, Microsoft Copilot: [Microsoft Copilot: the refreshed Copilot app](https://x.com/MSFTCopilot/status/2103457604571640263)
- 2026-09-25, Satya Nadella: [Satya Nadella: 'intelligence is accelerating fast. Now we need to diffuse it everywhere.'](https://x.com/satyanadella/status/2103475443089420408)
- 2026-09-25, Peter Steinberger 🦞: [Steinberger lists OpenClaw features contributed by Microsoft's Autopilot team](https://x.com/steipete/status/2103491221515907087)
- 2026-09-25, Sumit Chauhan: [Sumit Chauhan: bringing the full power of Microsoft Office into the Copilot app](https://x.com/sumit_c/status/2103486852833620405)
- 2026-09-25, Mustafa Suleyman: [Mustafa Suleyman: 'Autopilot is super cool'](https://x.com/mustafasuleyman/status/2103534705614397772)
- 2026-09-24, Alexis Conneau: [](https://x.com/alex_conneau/status/2103143665577423347)
- 2026-09-23, OpenAI: [OpenAI: ChatGPT Voice gets plugins, GPT-6 models and ChatGPT Work](https://x.com/OpenAI/status/2102808325742322002)
- 2026-09-21, Amazon News: [Amazon: $1.9B more for Delivery Service Partners; 20,000+ Smart Delivery Glasses by end of 2027](https://x.com/amazonnews/status/2102032303744491971)
- 2026-09-16, ElevenLabs: [](https://x.com/ElevenLabs/status/2100262886916358361)
- 2026-07-23, Claude: [](https://x.com/claudeai/status/2080376096873177300)
- 2026-05-20, ElevenLabs: [](https://x.com/ElevenLabs/status/2057155693623361667)