OpenAI DevDay 2026 Recap: Announcements

OpenAI DevDay 2026 took place on Tuesday, September 29, 2026 at Fort Mason in San Francisco, with Sam Altman headlining the free keynote at 10 a.m. Pacific — and the keynote is now behind us, so this page has flipped from live-along mode to the recap you can actually use. The headline: GPT-6.1 Sol, a near-Astra model at one-fifth of Astra's standard API price, plus a pile of agent infrastructure — Dots (always-on agents), Ultrafast speed tiers, a new Pro 500 plan, computer use in the Agents API, and managed OpenAI agents on AWS Bedrock.

Below you get both halves: the confirmed announcement list from the keynote, and the pre-event planning material — schedule with time-zone conversions, how to watch the recordings, and what the community expected going in — which is worth keeping for the DevDay Exchange events that start in October.

Quick answer

• What was announced: GPT-6.1 Sol (near-Astra intelligence at 1/5 Astra pricing, all paid tiers), Dots always-on agents, Ultrafast mode (up to 8× token speed in Codex), the Pro 500 plan, computer use + multi-agent workflows in the Agents API, Bedrock Managed Agents powered by OpenAI, Codex cloud + CLI refresh, and more — full list below.
• When it happened: Tuesday, September 29, 2026. Doors/breakfast 8:00 a.m. PT, opening keynote 10:00 a.m. PT, programming until the closing session at 4:00 p.m., reception to 7:00 p.m.
• Missed the keynote? The stream was free and breakout sessions land as recordings on openai.com shortly after the event.
• Biggest model takeaway: the expected wide Astra release did not happen — Astra gained an Ultrafast tier for top plans only, and GPT-6.1 Sol is the model most people actually get. Our GPT-6.1 Sol pricing and comparison post has the numbers.
• Biggest agent takeaway: the "managed agents" rumor was real, sort of — it shipped as Bedrock Managed Agents, powered by OpenAI, plus Dots for consumers.

OpenAI DevDay 2026 schedule (all times Pacific)

OpenAI's DevDay page lists the day like this:

Convert the 10:00 a.m. keynote to your own clock: 1:00 p.m. Eastern (EDT), 5:00 p.m. UTC, 6:00 p.m. London (BST), 7:00 p.m. Paris/Berlin (CEST), 10:30 p.m. India (IST), 1:30 a.m. September 30 in Tokyo (JST), and 1:30 a.m. September 30 in Beijing/Singapore (UTC+8). The breakout block runs through mid-afternoon Pacific, so even west-coast attendees get a full day; everyone east of Europe is into the night.

OpenAI has also been running DevDay Exchanges — smaller, regional one-day events — with stops announced for Bengaluru, Tokyo, Seoul, Paris, Berlin, London, São Paulo, and Mexico City later this year. If San Francisco wasn't an option, one of those is the practical way to attend in person.

How to watch the DevDay 2026 keynote live

The opening keynote is free to watch. OpenAI streams it and lets anyone sign up for a reminder on the DevDay page — no ticket, no application, no paywall. Two practical notes from past years that still apply:

1. The stream is the only same-day source. There is no day-of ticket sales, and in-person capacity went to people who applied months ago. If a "watch party" link slides into your feed this morning, treat it as a scam — the official stream is the one OpenAI links from its own page.
2. Everything else lands as recordings. OpenAI posts the breakout sessions on openai.com shortly after the event, so the parallel-track talks you have to miss while asleep (that's most of Asia and Europe for a 10 a.m. Pacific keynote) are recoverable the same day.

If you're catching up in text instead of video, the ecosystem live blogs and live-update threads from developer media captured the keynote as it landed, and OpenAI's own recap page consolidates the official list. Everything above in the announcements section reflects the keynote as delivered.

Everything announced at DevDay 2026 (keynote recap)

OpenAI's own DevDay 2026 recap runs 25+ items long. These are the ones that matter to developers, with what was actually said:

Models and pricing

• GPT-6.1 Sol — near-Astra performance for complex coding, computer use, and professional work at one-fifth of Astra's standard API prices ($2/$10 per million tokens). Available to every paid tier from day one, including the API, Plus, Pro, Business, Enterprise, and Edu. Full rates, context specs, and comparisons in our GPT-6.1 Sol post.
• Ultrafast mode — a new speed tier pushing up to 300 output tokens per second in Codex (8×) and 6× via the API. Live today for GPT-6 Astra on the Pro 500 and Enterprise plans ($60/$300 per million tokens); Sol Ultrafast is listed as coming soon.
• Pro 500 — a new subscription tier with 25× Plus usage that includes Ultrafast access.
• Astra's "wide release" didn't happen — Astra kept its September 3 pricing ($10/$50) and general availability; the Ultrafast tier is the only new access path, and it's limited to the top plans.

Agents

• Dots — OpenAI's new always-on agents that "handle everything" in the background, available on Pro and Business Premium (Enterprise, Edu, and Healthcare in beta).
• Agents API upgrades — computer use, Codex multi-agent workflows, tool search, and context compaction, with OpenAI hosting the infrastructure.
• Bedrock Managed Agents, powered by OpenAI — the AWS partnership the late-September reporting hinted at. OpenAI's agent stack now runs on Amazon Bedrock; the "managed agents" rumor was real, just under a co-brand name.
• Decisions API — a Luna-based API for real-time classification and decisioning, in limited preview with wider access "in days."

Codex

• Codex in the cloud — reusable development environments you can start from any device, rolling out across Plus, Pro, and Business plans.
• Codex CLI refresh — voice control, a new /agents view for delegating multiple tasks, plus prompt editing, session resume, and worktree improvements across all plans.
• Code Review — automatic cloud review from ChatGPT's desktop app and directly on GitHub pull requests and GitLab merge requests.
• Codex Security Cloud — on-demand and scheduled repository scanning, finding triage, and cloud fixes, powered by a new model called Daybreak Blue.

Platform

• Private Intelligence — zero-data-retention processing via Private Safety Processing, with Private Inference entering preview this fall.
• ChatGPT platform — plugin extensions, ChatGPT Space and Pages, collaborative slides, @ChatGPT in Slack and Teams, a Meetings plugin for macOS (beta), and shareable profiles.
• Sign in with ChatGPT — first 16 partners including Devin, Notion, and Vercel, with plan usage that follows you across partners.
• OpenAI Marketplace — 32 partners at launch; the keynote called out CodeRabbit, Notion, and Vercel.
• DevDay Exchange — the ten-city world tour (Bengaluru, Tokyo, Seoul, Paris, Berlin, London, São Paulo, Mexico City among them) starts in October.

One more number from the keynote: OpenAI now claims 1.2 billion weekly users — worth remembering when you read "niche" next to any of the above.

Can you still attend DevDay 2026 in person?

No — and that's not a sales pitch, it's arithmetic. Attendance required an application; applications closed on July 10, 2026; in-person tickets cost $650; the DevDay page explicitly says there are no day-of tickets and tickets are non-transferable. Scholarships existed alongside the application window, and that window has closed too. Your realistic options today: watch the free keynote stream, catch the breakout recordings afterward, or plan for one of the eight DevDay Exchange cities later this year.

FAQ

What time is the OpenAI DevDay 2026 keynote?

10:00 a.m. Pacific Time on Tuesday, September 29, 2026 — Sam Altman's opening keynote, per OpenAI's own schedule page. That's 1 p.m. Eastern, 5 p.m. UTC/London-adjacent (6 p.m. BST), 7 p.m. in Paris and Berlin, 10:30 p.m. in India, and 1:30 a.m. on September 30 for Tokyo and Beijing. Breakfast opens at 8 a.m. PT and the reception ends at 7 p.m. PT.

What is the full DevDay 2026 schedule?

Breakfast at 8:00 a.m. PT, opening keynote at 10:00 a.m., breakouts and programming from 11:15 a.m. to 3:30 p.m. (with lunch at 11:30), closing session at 4:00 p.m., and a reception from 4:45 to 7:00 p.m. — all Pacific Time, all at Fort Mason in San Francisco.

Is OpenAI DevDay free to watch?

The opening keynote livestream is free and open to everyone — you can sign up for a reminder on OpenAI's DevDay page. In-person attendance is the paid, application-only part ($650). Breakout sessions aren't streamed live but are recorded and posted on openai.com shortly after the event.

Can I buy a DevDay 2026 ticket today?

No. Applications closed July 10, 2026, there are no day-of tickets, and tickets are non-transferable. For in-person events, look at the DevDay Exchanges announced for Bengaluru, Tokyo, Seoul, Paris, Berlin, London, São Paulo, and Mexico City.

Where is OpenAI DevDay 2026 held?

Fort Mason in San Francisco. The venue hosts the keynote, the breakout programming, and the evening reception.

Did OpenAI announce managed agents at DevDay?

Sort of — under a co-brand name. The keynote confirmed Bedrock Managed Agents, powered by OpenAI: OpenAI's agent stack, running on Amazon Bedrock through a partnership with AWS. OpenAI's own consumer-facing answer to always-on agents is Dots, available on Pro and Business Premium. The late-September rumor was real; the product names just weren't what anyone guessed.

What did OpenAI announce at DevDay 2026?

The model news: GPT-6.1 Sol, a near-Astra model at one-fifth of Astra's standard price, available to all paid tiers, plus the Ultrafast speed tier (Astra-only for now) and a new Pro 500 plan. The agent news: Dots always-on agents, computer use and multi-agent workflows in the Agents API, and Bedrock Managed Agents powered by OpenAI. The Codex news: Codex in the cloud, a CLI refresh with voice and an /agents view, Code Review on GitHub PRs and GitLab MRs, and Codex Security Cloud. Platform news included Private Intelligence, ChatGPT plugin extensions, and Sign in with ChatGPT's first 16 partners. The expected wide Astra release did not materialize — Astra kept its existing pricing and availability.