OpenAI DevDay 2026: dots, GPT-6.1 Sol and a $500 Pro tier
OpenAI announced more than 20 updates. Four change what you should pay for or build with. Here they are first, then the full list, the leaks we scored, and our read.
OpenAI’s personal agent is here—but only on its top plans. And its new mid-tier model costs exactly what Sonnet 5.5 costs.
dots are always-on agents with their own computers, included (one each) in Pro and Business Premium. GPT-6.1 Sol lists at $2 / $10 per million tokens. The new $500 Pro tier adds the fastest Astra.
OpenAI’s DevDay recap (opens in a new tab)Four launches that change a decision
dots
Always-on agents powered by GPT-6 Astra, each with its own cloud computer and browser. Give one a goal and rules for what it may do alone; it works in the background, messages you in ChatGPT, SMS, Slack or Teams, and asks before sensitive actions.
Who gets it: Rolling out in eligible markets to Pro and Business Premium (first Dot included in the plan); Enterprise beta when an admin enables it. More Dots, or faster Dots, will be a fixed monthly add-on later.
Cost: Chatting with your Dot does not count against ChatGPT limits; tasks it starts in Codex or ChatGPT Work do.
OpenAI announcement (opens in a new tab)GPT-6.1 Sol
A major upgrade to Sol that OpenAI says comes close to GPT-6 Astra on hard coding, computer-use and professional tasks.
Who gets it: Plus, Pro, Business, Enterprise and Edu in ChatGPT Work and Codex (not in chat yet); API as gpt-6.1-sol.
Cost: API: $2 input / $0.10 cached input / $10 output per 1M tokens—the same list price as Claude Sonnet 5.5.
OpenAI announcement (opens in a new tab)Sign in with ChatGPT—and bring your plan
Sign in to other apps with your ChatGPT account. Plus and Pro users can also let an app spend their plan’s ChatGPT Work and Codex usage for its AI requests—no API key—and cap each app’s weekly share in Settings > Usage.
Who gets it: Plan usage in 12 apps at launch, including Devin, Warp, Vercel, Notion, Amp, Conductor, Kilo Code and Hermes Agent (Lovable coming); sign-in only in Airtable, Canva, GitLab, HubSpot and Supabase; open-source integrations include OpenClaw, OpenCode, Pi and T3.
Cost: Counts against your existing plan; it adds no new allowance, and apps may charge their own fees. Credits after the limit are off by default.
OpenAI announcement (opens in a new tab)Pro $500 and Astra Ultrafast
A new $500/month Pro tier with the highest usage limits and exclusive Astra Ultrafast in ChatGPT Work and Codex (up to 8× standard Astra speed; up to 6× in the API).
Who gets it: New Pro tier; Ultrafast also in the API.
Cost: $500 per month. Separately, the $200 Pro reopens on September 30 with about half the old API-dollar usage.
OpenAI announcement (opens in a new tab)The full list, grouped
Work with agents
- dots (always-on agents)
- ChatGPT Space: shared workspaces with ChatGPT and your team (Pro, Business, Enterprise)
- Live pages and collaborative slides (Pro, Business, Enterprise)
- Team tasks, @ChatGPT in Slack and Teams (Business, Enterprise)
- Meeting plugin, beta for Pro and Business
- Shareable profile pages
Build with Codex and the API
- GPT-6.1 Sol
- Astra Ultrafast and Sol Ultrafast
- Codex Cloud (run anywhere, reusable environments)
- Codex CLI with voice and an /agents view
- Code review in the ChatGPT desktop app
- Codex Security Cloud
- Decisions API (limited preview)
- Computer use in the Agents API; Amazon Bedrock managed agents
- Private intelligence: zero data retention with safe processing, private inference preview
Platform and subscription
- Plugin extensions, better plugin submission and discovery
- Plugins on ChatGPT Sites (Business and up; Sites itself is a public beta for Plus and up); MCP event-driven automations
- Sign in with ChatGPT
- New $500 Pro tier
- OpenAI software marketplace for eligible enterprises (32 launch partners)
OpenAI’s recap says GPT-6.1 Sol costs a quarter of Astra; its own Sol announcement says a fifth and lists $2 / $10. We use the listed prices.
What testers found in the first two days
dots
Dan Shipper, Every (Sep 29)
Too buggy to recommend yet: frequent permission problems, dropped messages, and a Dot that often could not reach its in-app browser. He still found himself reaching for his Dot instead of opening new Codex threads, and expects fixes within a week or two.
Read the source (opens in a new tab)dots
Corey Noles, The Neuron (Sep 30)
About five minutes to set up, with an interview-style onboarding. The Dot spotted a loose end on its own (a warranty email) and offered to help—but it also sent an email before he had reviewed it, until he told it to show outgoing messages first. Not available in the EEA, the UK or Switzerland.
Read the source (opens in a new tab)ChatGPT Space
Dan Shipper, Every (Sep 29)
The smoothest part of the launch for him: tag your Dot in a comment on a document and get the revision inline while you keep working.
Read the source (opens in a new tab)Decisions API
Every (Sep 29)
On a replay of a text-based computer task, Decisions picked the right control in 76 of 78 scored steps versus 73 for Jev, answering in about 230 ms versus 500 ms. One task, one tester—a promising sign, not a benchmark.
Read the source (opens in a new tab)GPT-6.1 Sol
Artificial Analysis (checked Oct 1)
52 on its Intelligence Index, 10th of 222 models, at about 66 output tokens per second—slower than average—and $0.72 per task in their run.
Read the source (opens in a new tab)GPT-6.1 Sol
Handy AI roundup (Sep 29)
Relief that it fixes GPT-6 Sol’s weak coding; several developers called the cheaper cached input “the actual big announcement” for long agent loops. One flagged that Sol’s own system card shows a higher coding-deception rate than its predecessor.
Read the source (opens in a new tab)Early-access and day-one reports describe one person’s setup. We label each by kind and link the original.
What the pre-event leaks got right
An always-on personal agent
TestingCatalog (Sep 26), from ChatGPT configuration strings
dots are always-on agents with their own identity and cloud computer.
Leak source (opens in a new tab)It would be called “O” (internal project “Aeon”)
Same report
The product is called dots.
Leak source (opens in a new tab)The name could be “dots”
Readers’ reading of the “on the dot” teaser
A guess from a pun that turned out correct—still a guess until OpenAI said it.
Leak source (opens in a new tab)GPT-6.1 Astra will not ship
Several outlets, citing the Wall Street Journal, reported the day before DevDay that OpenAI shelved it over safety testing. OpenAI’s recap does not mention it; GPT-6 Astra remains its most capable model and now powers dots.
Report (opens in a new tab)What it means for you
Your ChatGPT plan becomes a wallet
If Devin, Warp or Vercel can bill your Plus or Pro plan, startups can ship AI features without passing API costs to users—and OpenAI gets to sit between those apps and their customers.
Personal agents are now a paid-plan feature
OpenAI puts the first Dot inside Pro and Business Premium, while Manus’s Cue launched free with invite codes. Expect agents to become the reason to pick one subscription over another.
A mid-tier price war
GPT-6.1 Sol lists at $2 / $10 per million tokens—exactly Claude Sonnet 5.5’s price, one day later. Run your own task on both before switching.
Watch the usage fine print
Talking to a Dot is free of limits; the work it starts in Codex or ChatGPT Work is not. An agent that works while you sleep can spend while you sleep.
Questions readers ask
What are OpenAI dots?
Always-on agents powered by GPT-6 Astra. Each has its own cloud computer and browser, connects to apps you choose, works in the background and asks for approval before sensitive actions.
Source (opens in a new tab)Who can use dots and how much do they cost?
They are rolling out in eligible markets to ChatGPT Pro and Business Premium, where the first Dot is included in the plan; Enterprise can trial a beta. OpenAI says more Dots will be a fixed monthly add-on later.
Source (opens in a new tab)How much does GPT-6.1 Sol cost?
In the API, $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. In ChatGPT it is available in ChatGPT Work and Codex on paid plans.
Source (opens in a new tab)Is GPT-6.1 Astra cancelled?
News outlets citing the Wall Street Journal reported it was shelved. OpenAI’s DevDay recap does not mention it.
Source (opens in a new tab)October 1, 2026 · keynote plus two days of hands-on reports. Official launches, press reports, scored leaks and our opinion are labelled separately. Availability is rolling out by market and plan; check your own account.