Live from OpenAI DevDay 2026: Keynote

Here is the catalogued entry for the video:

Summary

This video is the official keynote presentation from OpenAI DevDay 2026, hosted at Fort Mason in San Francisco. Presented primarily by OpenAI CEO Sam Altman, alongside product manager Holly Li, research lead Tejal Patwardhan, and Romain Huet, the keynote announces major product launches and research milestones across OpenAI's ecosystem. The primary announcements include the "dots" always-on agent platform, ChatGPT Space, GPT-6.1 Sol, the Ultrafast inference tier, Codex in the Cloud, Codex Security Cloud, and the Agents and Decisions APIs.

---

What is shown

• [00:16] Sam Altman opens the keynote, reviewing user requests shipped since the last DevDay, including Codex for Linux, multi-folder project support, and Codex mobile.
• [02:26] Introduction of dots, autonomous always-on AI agents that manage proactive tasks, with a promo video showcasing users interacting with personalized agents (Alfred, Felipe, Jojo) across phone, desktop, and dedicated home display terminals.
• [08:21] Announcement of ChatGPT Space, a collaborative workspace uniting human teammates, living documents, and AI agents.
• [09:27] Live demo by Holly Li showing her dot agent ("Dottie") helping coordinate a launch for a fictional app called "Blossom Music", generating charts, drafting FAQs, and deploying an updated UI to Codex and an iOS simulator.
• [19:40] Launch of GPT-6.1 Sol, a cost-effective frontier model offering near-Astra capabilities.
• [20:34] Announcement of the Ultrafast inference tier, demonstrated with a side-by-side comparison of 3D asset generation (a rocket launching) running at standard vs. Ultrafast speeds.
• [21:13] Preview of the Decisions API, showcasing near-instantaneous flight booking UI execution using the Luna model.
• [24:11] Research segment presented by Tejal Patwardhan on using models as automated research interns, computer-use harness optimization, and safety stress test benchmark results.
• [27:42] Updates to the developer platform: open-sourcing the Codex Harness, launching Codex in the Cloud, Codex Security Cloud, and the Agents API (featuring GPT-6 Astra computer use).
• [31:42] Live demo by Romain Huet: creating and manipulating a 3D simulation of Fort Mason with dots using Ultrafast, building an attendee ticket giveaway web app via the Codex CLI, controlling a 3D spaceship game ("Astra Adventures"), and interacting with Hugging Face's physical programmable robot ("Microduck / Lavender").
• [42:28] Launch of Sign in with ChatGPT for partner applications, Plugin Extensions, and the OpenAI Marketplace.
• [48:45] Presentation of an OpenAI community tribute film and announcement of the limited-edition handheld console for attendees, followed by a live countdown button reset by community member Tibo.

---

Claims & numbers

• Scale & Reach: Sam Altman claims ChatGPT currently has 1.2 billion weekly active users, over 40 model launches have occurred since the last DevDay, and 2.5 million businesses are built on OpenAI.
• ChatGPT Space & Ecosystem: Over 4,000 plugins/apps are integrated and accessible via ChatGPT.
• GPT-6.1 Sol Pricing & Benchmarks: Altman states that GPT-6.1 Sol provides near-Astra intelligence at one-fifth the price: $2.00/M input tokens ($0.10/M cached input, a 95% discount) and $10.00/M output tokens (compared to GPT-6 Astra at $10 input / $1 cached / $50 output).
• Ultrafast Performance & Tier: Ultrafast delivers 8x standard speed (up to 300 tokens/second) at 6x standard cost. It is included in the new Pro 500 ($500/month) plan, which features 25x the usage limits of Plus with no 5-hour limit. Pro 200 subscriptions are also reopened.
• Research & Autonomous Milestones: Altman and Patwardhan state OpenAI reached its goal of having an automated AI research intern by September 2026. Patwardhan reports that agent usage in research skyrocketed over summer 2026, and the success rate on long-horizon research tasks (8–16 hours with zero human interventions) increased from 10% in January 2026 to 35% in July 2026.
• Safety Benchmark: Patwardhan presents a computer-use safety stress test where GPT-6 Astra achieved a 2.4% misaligned outcome rate, outperforming Claude Opus 5.5 (6.1%) and Claude Fable 5.1 (9.5%).
• Platform & API Reliability: Responses API usage experienced 100x growth over the past year while maintaining >99.9% uptime; tool-call workflows are sped up by >30%, and time-to-first-token is reduced by >45%.
• OpenAI Marketplace: Launched with 30+ launch partners, allowing enterprise clients to use existing OpenAI contractual commitments toward third-party tools.

---

Notable quotes

• [01:31] "This is the real-deal version of AI that we've always imagined for ourselves. It's like an AI helper that always has your back, inspired by the cool versions of what we all watched in movies growing up." — Sam Altman
• [20:56] "An incredible 300 tokens per second. You know what, it's worth it." — Sam Altman
• [52:52] "I believe that the future, if we get this right, can be much more like a new Renaissance than a new Industrial Revolution." — Sam Altman

---

Assessment

This is an authentic, full-length live presentation of the OpenAI DevDay 2026 Keynote. The presentation combines onstage speeches, recorded concept/marketing videos, live software and hardware interactions, and benchmark metric presentations. While the prepared video packages illustrate envisioned real-world integration, the onstage live demos (notably Romain Huet's interactive sessions) run in real time and openly experience minor live glitches (e.g., initial voice connection timeout at 31:18, smoothly recovered via CLI).

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.