OpenAI DevDay 2026

Summary
In this livestream, Matthew Miller (founder of BridgeMind) reacts to and streams the official OpenAI DevDay 2026 keynote, while simultaneously running multiple AI agent workflows in BridgeMind and BridgeGame.ai. After the keynote—which introduced “dots” agents, ChatGPT Space, GPT-6.1 Sol, the UltraFast tier, and a $500/month Pro plan—Matthew subscribes to the new $500 tier and tests GPT-6.1 Sol and GPT-6 Astra UltraFast on benchmarks, 3D asset generation in Blender, and browser-based web games.

What is shown

• Pre-Keynote BridgeMind Workflows [00:00–144:00]: Matthew coordinates multiple agent threads running Claude Opus 5.5, Claude Sonnet 5.5, and GPT-6 Astra to build games on BridgeGame.ai (poker, an FPS gauntlet game), while discussing OpenAI pricing tier adjustments.
• OpenAI DevDay 2026 Keynote Broadcast [144:19–197:25]:
  • 144:19: Sam Altman opens DevDay 2026, recapping growth milestones and product updates including Codex on Linux and Codex Mobile.
  • 146:21: Announcement of dots, persistent autonomous agents running 24/7 in the cloud powered by GPT-6 Astra.
  • 153:25: Holly Li demos “Dottie,” showing mobile/desktop handoffs, voice calls, calendar scheduling, Slack bug reporting, and background code execution for a fictional app called “Blossom Music.”
  • 152:16: Introduction of ChatGPT Space, collaborative living documents and presentations integrating dots and human teams.
  • 163:35: Announcement of GPT-6.1 Sol, designed to offer near-Astra intelligence at one-fifth the cost.
  • 164:30: Announcement of UltraFast, delivering generation speeds up to 300 tokens per second (8x standard speed).
  • 165:32: Introduction of the Pro 500 ($500/month) plan with UltraFast access, no 5-hour limit, and 25x Plus usage.
  • 166:10: Announcement of Decisions API for low-latency, sub-second routing and classification using the Luna model.
  • 168:14: Tejal Patwardhan discusses research acceleration, safety stress testing, and agentic computer-use benchmarks.
  • 175:38: Romain Huet demos Codex CLI updates, 3D world editing in real-time, browser automation in a space game (“Astra Adventures”), and connects multimodal APIs to Lavender, a Hugging Face microduck robot.
  • 188:25: Distribution features unveiled: “Sign in with ChatGPT”, full-app Plugin extensions, and the OpenAI Marketplace.
• Post-Keynote Benchmarking and Testing [198:00–318:00]:
  • Matthew reviews Artificial Analysis leaderboard rankings for GPT-6.1 Sol.
  • Runs comparative evaluations on BridgeBench Design Gallery (Lava Lamp, Sunset Ocean, Black Hole Merger, Rocket Launch) comparing GPT-6.1 Sol against GPT-6 Sol, GPT-6 Astra, and Claude Opus 5.5.
  • Upgrades his account to the $500/month ChatGPT Pro plan [247:15].
  • Tests UltraFast with GPT-6 Astra in Blender via MCP to generate a detailed 3D rocket model [250:54].
  • Tests a multiplayer browser FPS zombie game (“Blackwater Relay”) generated by GPT-6.1 Sol on BridgeGame.ai [269:05, 311:00].
  • Inspects Codex token/rate-limit consumption, noting his weekly limit dropped rapidly while using UltraFast [257:40, 289:30].

Claims & numbers

• OpenAI Keynote Claims:
  • ChatGPT has 1.2 billion weekly active users and 2.5 million business customers [144:30, 188:05].
  • Over 40 model launches took place over the preceding year [144:30].
  • GPT-6.1 Sol Pricing: $2.00 / 1M input tokens ($0.10 cached) and $10.00 / 1M output tokens, advertised at 1/5th the price of GPT-6 Astra ($10 / $50) with 95% input cache discount [164:00].
  • UltraFast Tier: Up to 300 tokens/second, 8x faster than standard, priced at 6x standard price [164:50].
  • Pro 500 Tier: $500/month, includes UltraFast access, no 5-hour limit, 25x usage allowance of the Plus tier [165:40].
  • Responses API latency cut time-to-first-token by 45% and sped up tool calling by 30%+ with 99.9%+ uptime [174:50].
  • In internal research tasks, agent usage grew to take on over one-third of day-long tasks with zero interventions by July 2026 [170:30].
• BridgeMind / Matthew’s Claims:
  • BridgeMind reached over $255,000 ARR during the stream, supported by a 65% off promotional coupon [107:30, 224:50].
  • Matthew states that using GPT-6 Astra with UltraFast consumed 2% of his total weekly limit within a single 2-minute prompt [257:40], and reached 0% remaining weekly quota in under 30 minutes of agent work [289:30].

Notable quotes

• “Today we’re launching UltraFast across the API, ChatGPT, and Codex... an incredible 300 tokens per second.” — Sam Altman [164:30]
• “We are introducing Pro 500 with our highest usage limits and access to UltraFast in ChatGPT and Codex.” — Sam Altman [165:32]
• “I ran out of my entire weekly limit on the $500 ChatGPT Pro Max plan in under 30 minutes... GPT-6 Astra UltraFast.” — Matthew Miller [289:27]

Assessment
This video is an authentic community livestream by developer Matthew Miller reacting live to the official OpenAI DevDay 2026 keynote stream, followed by unedited, real-time testing of the newly announced models and tiers. All software interactions, terminal commands, web app deployments, and subscription purchases occur live on screen, documenting real performance, rapid rate-limit consumption, and actual API benchmark comparisons.

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