OpenAI packed more than 20 launches into one keynote at DevDay 2026, headlined by a cheaper flagship model and a new class of agents that keep working after you close the tab. Here's what actually shipped, what it costs, and what it changes for anyone using ChatGPT or the API.
OpenAI held its DevDay 2026 keynote in San Francisco on September 29, and the company's own recap calls it the biggest one yet, with more than 20 announcements spread across ChatGPT, Codex, and its model lineup. The headline items are GPT-6.1 Sol, a mid-tier model that OpenAI says lands close to its top model at a fraction of the price, and Dots, always-on agents that run in the background on your behalf.
A packed hour. A mixed bag.
Not all of it lands the same way. What actually changes for you?
There's a lot to sort through. Some of it amounts to a genuine price change that affects what developers pay tomorrow. Other pieces are a first look at features that won't reach most people for weeks. The rest is plumbing aimed at teams building on OpenAI's platform. This is a look at the items that change how the products actually work.
What OpenAI announced at DevDay 2026
The keynote split into three buckets: a new model and speed tier, a new way to run agents, and a rebuilt developer surface. Below is the short version of what landed and who gets it first.
Category | What's new |
|---|---|
Model | GPT-6.1 Sol, a mid-tier model with near-Astra performance at a lower price |
Speed tier | Ultrafast, up to 300 tokens per second for interactive work |
Agents | Dots, always-on agents with their own cloud computer and access to 4,000+ apps |
Developer tools | Agents API in public beta, Decisions API in limited preview, cloud-hosted Codex |
Collaboration | ChatGPT Space, Pages, and shared tasks for teams |
Accounts | Sign in with ChatGPT, so a user's plan can cover usage inside partner apps |
Plans | Pro 500, a new top consumer tier with 25 times the Plus allowance |
The recap frames all of this around one number: 1.2 billion weekly ChatGPT users. Nearly every launch is pitched as a way to open that audience to developers and give existing users more to do inside the app.
GPT-6.1 Sol: near-Astra intelligence at a lower price
GPT-6.1 Sol is the model announcement that matters most for developers. OpenAI describes it as a major upgrade to GPT-6 Sol from a week earlier, with strong results on agentic coding, computer use, and professional work. The company claims it nearly matches the intelligence of its top model, GPT-6 Astra, at one-fifth of Astra's standard token price.
The numbers back that up on price. GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens, the same list price GPT-6 Sol launched at. The real cut is cached input, which drops to $0.10 per million, a 95% discount from standard input pricing.
That cache rate is the part developers have been talking about. If your app resends a long system prompt and tool list on every turn, the cached price is where your bill goes down. On Hacker News, one commenter called it "the actual big announcement."
Independent checks mostly line up. Artificial Analysis put Sol at 52 on its Intelligence Index, one point under Astra's 53, with a cost per task around 22% of Astra's. Anthropic's Claude Opus 5.5 still tops that index at 58, so Sol isn't the new leader on raw capability. It's competitive, and it's far cheaper.
GPT-6.1 Sol is available now through the API, and to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. It hasn't reached the standard ChatGPT chat interface yet.
Ultrafast and the Pro 500 plan
Ultrafast is OpenAI's new speed tier. It generates tokens up to eight times faster than standard in Codex, about 300 tokens per second, and up to six times faster in the API. That speed costs six times the standard rate, running $60 per million input tokens and $300 per million output for Astra.
When does paying that premium make sense? It comes down to whether a person is waiting on the model. Pair-programming sessions in Codex, where you watch suggestions appear, are the clearest fit. For batch jobs that run unattended, Sol at standard speed with a warm cache works out far cheaper per output token.
Ultrafast in ChatGPT requires the new Pro 500 plan, which OpenAI says carries 25 times the usage allowance of ChatGPT Plus. The company also reopened Pro 200 on stage, the tier whose new sign-ups it had paused in September. GPT-6.1 Sol Ultrafast is listed as coming soon.
Dots: agents that keep running after you log off
Dots are the flashiest launch. A dot is an agent that keeps working when you close the tab. OpenAI says each one is powered by GPT-6 Astra, gets its own cloud computer, learns from feedback, and can pursue a goal around the clock. Dots connect to more than 4,000 apps through plugins, live inside ChatGPT, Slack, and Teams, and can take voice calls.
The pricing is friendlier than it sounds. Your first dot is included in a Pro or Business Premium plan at no extra cost, and conversations with it don't count against your ChatGPT usage limits. Businesses get specialist dots with their own identity and credentials, managed through Microsoft's admin tools.
OpenAI's safety page sets the guardrails. Dots use read-only tools for proactive research, get auto-reviewed, and hand sensitive tasks like changing a password back to you. The same page warns that "dots can still make mistakes," which is worth keeping in mind before you delegate anything consequential.
For everyday users, the practical question is what you'd actually hand off. Think monitoring a topic and sending you a summary, or chasing an invoice without you reopening the app. It's a shift from asking a chatbot a question to giving it a standing job.
A rebuilt platform for developers
The developer side of DevDay got the most under-the-hood changes. The Agents API moves Codex's multi-agent loop into a hosted service, bundling tool search, tool calling, and context management so you don't have to build and maintain that scaffolding yourself. It's in public beta, and computer use is part of the package.
Codex also moved into the cloud. Tasks now run remotely, so you can start a job on your phone and pick up the result on any device. The refreshed Codex CLI can start and steer tasks by voice and adds an /agents view for tracking them. Code Review handles GitHub pull requests and GitLab merge requests, and Codex Security Cloud scans entire repositories on demand or on a schedule.
One more item deserves a mention: the Decisions API. It points a smaller model, Luna, at a fixed set of questions with a finite set of answers, which fits tasks like routing a request or labeling a ticket. It's in limited preview, with a wider release promised shortly.
Sign in with ChatGPT and the partner question
Sign in with ChatGPT lets a user log into a partner app with their ChatGPT account, and eligible usage counts toward their existing plan limits. The catch is the partner count. OpenAI's official recap names 16 launch partners, including Cognition's Devin, Notion, Vercel, T3, and OpenClaw. Codex lead Tibo Sottiaux said "over 60" on X. Both numbers are out there, and they don't match, so treat the larger figure as a forward-looking claim rather than a launch-day fact.
For developers, the model is a way to remove a signup step and let someone try your app without a new credit card. For OpenAI, it pulls more third-party usage into its billing. Whether the count lands closer to 16 or 60 is worth watching as the program expands.
What the DevDay 2026 launches mean for users
The through-line is that OpenAI is turning ChatGPT from a place you ask questions into a place where agents and apps run. Dots, Space, Pages, and the plugin platform all point the same direction: more work happening inside the product, on OpenAI's infrastructure.
Big pivot.
For most people, the near-term changes are small. GPT-6.1 Sol is cheaper for developers but not yet in the chat interface. Ultrafast needs a top-tier plan. Dots are gated by plan and region. The features that touch everyone, like the usage reset announced live, come and go fast.
Small shifts, mostly. The bigger ones take time to reach you.
Developers should look hardest at the pricing. A 95% cut on cached input quietly changes what an agent that resends a long prompt costs to run, and that's the kind of thing that shows up in a monthly bill rather than a keynote slide.
GPT-6.1 Sol is live in the API and in ChatGPT Work and Codex now. Check OpenAI's DevDay recap and pricing page for the specifics on your plan, since availability for Dots and Ultrafast varies by account and region as it rolls out.
