What OpenAI Announced at DevDay 2026: dots, GPT-6.1 Sol, Ultrafast, the $500 Pro Tier — Who Gets What, and How to Pay (2026)
On September 29, 2026, OpenAI announced dots, GPT-6.1 Sol, Ultrafast models, Codex Cloud, the Decisions API, computer use in the Agents API and a new $500-per-month Pro tier. Here is what each one is, which ChatGPT plans and API users get it, and the two practical routes if your card is declined or you need an OpenAI-compatible endpoint.
1. The short answer: what DevDay 2026 announced
OpenAI held DevDay 2026 on September 29, 2026. Its recap describes the event as the largest so far, with more than 20 major updates across ChatGPT, Codex and its models, plus new ways of working with AI. OpenAI also says it is opening ChatGPT up as a shared platform for people and agents, so developers can ship new native experiences directly to its combined 1.2 billion users. The announcements group neatly into five tracks. Models: GPT-6.1 Sol and the Ultrafast variants. ChatGPT: the dots agents, a new $500-per-month Pro tier, and new collaboration features: a space for your materials, dynamic pages and collaborative slides. Codex: Codex Cloud, code review, Codex Security Cloud, and a Codex CLI with voice conversation, an /agents view and built-in worktrees. API: the Decisions API and computer use in the Agents API. Enterprise: zero data retention, a private reasoning preview, and an OpenAI software marketplace, in beta, where eligible enterprise customers can put part of their existing OpenAI committed spend toward approved partner software, with 32 partners at launch. There were also plugin expansions, sites built with plugins, event-driven plugin automations via MCP, and partner products for Sign in with ChatGPT. On the timeline, DevDay followed GPT-6 Astra on September 3 and GPT-6 Sol and Luna on September 22, so the event was largely about bringing the GPT-6 generation into products.
2. What dots are and who gets them
dots are the new agents on the ChatGPT side. According to the recap, they are powered by GPT-6 Astra, which OpenAI calls its most intelligent and best-aligned model; each has its own cloud computer; and they work across the apps you choose to connect. In plain terms, this is not a smarter chat box but an agent that gets work done on a computer in the cloud, inside the apps you authorize. Availability is explicit: Pro, Business Premium and Enterprise. Plus is not on the list, and the recap does not mention an API for dots. One practical note: what a dot can do depends on which apps you connect, so think about what data it needs before granting access, and on company accounts check your administrator's policy first.
3. GPT-6.1 Sol: capability, pricing, and why it is not in chat yet
GPT-6.1 Sol is an upgrade to GPT-6 Sol. OpenAI says it is especially strong at agentic coding, computer use and professional work, bringing Sol closer to GPT-6 Astra on hard tasks at one fifth of Astra's standard input and output token prices. OpenAI also stresses that GPT-6 Astra remains its most capable frontier model overall. Some of the official comparisons: on the DeepSWE v1.1 coding eval, it performs comparably to Astra at roughly one fifth of the cost and beats GPT-6 Sol's best score by 6.4 points; on OSWorld 2.0 for computer use, at the highest reasoning setting it is 7 points above GPT-6 Sol and within 2.1 points of Astra, at about one seventh of Astra's cost per task; on Terminal-Bench Science 0.1, at the highest reasoning setting its average cost per task is $5.47 against $23.80 for Astra, but Astra still ranks first at 68.1%, and OpenAI recommends Astra for the most demanding scientific research. On factual errors at the extra-high (xhigh) reasoning setting, GPT-6.1 Sol scores 4.1%, versus 4.5% for GPT-6 Sol and 4.0% for Astra — OpenAI notes these were deliberately difficult prompts and do not reflect typical use. There is one availability detail people miss: Plus, Pro, Business, Enterprise and Edu users can use it from launch day in ChatGPT Work and Codex, but OpenAI states that these models are not yet available in chat. If you cannot find it in the regular chat model picker, that is expected. The Codex models page in OpenAI's ChatGPT help documentation (learn.chatgpt.com) also describes a staged rollout: Enterprise and Edu keep it off by default until an administrator enables it, and Free and Go are not included at launch, so go by what your own account shows. In the API, the model ID is gpt-6.1-sol. The line-by-line price breakdown, three worked cost examples and what to do when payment is declined are at /hub/gpt-61-sol-api-china.
4. Ultrafast and the new $500-per-month Pro tier
Ultrafast is the speed track. Per the recap, it raises token generation speed in Codex by up to 8x (300 tokens per second) and in the API by up to 6x. GPT-6 Astra Ultrafast is live in the API and available in ChatGPT Work and Codex on Pro $500 and eligible Enterprise and Edu plans; OpenAI calls it the world's fastest frontier model, up to 8x faster than GPT-6 Astra. GPT-6.1 Sol Ultrafast was announced alongside it in the GPT-6.1 Sol post, while the DevDay recap lists it as coming soon — check OpenAI's model pages for its current status. Tied to Ultrafast is a new Pro tier at $500 per month. OpenAI says it exists for heavier building needs: it offers the highest usage limits and access to Astra Ultrafast in ChatGPT Work and Codex, and computer use from the Agents API is also open to this tier in Codex and ChatGPT Work. For the price and benefits of other ChatGPT plans, check OpenAI's ChatGPT pricing page rather than relying on assumptions. Whether the new tier is worth it comes down to the usual test: do you actually hit usage limits every day, and do you need Ultrafast speed — not how loud the launch was. How to choose between the $200 and $500 Pro tiers, and whether either pays, is covered at /hub/chatgpt-pro.
5. What changed in Codex
Codex got four updates: Codex CLI and code review are labeled all plans, and Codex Security Cloud is for all Codex users. (1) Codex Cloud runs Codex on demand: on your own computer, remotely from your phone, or in the cloud from any device. The recap names Plus, Pro, Business, Healthcare, Education and Enterprise, then also labels it all plans, so go by what your account shows. (2) Codex CLI adds voice conversation, a new /agents view and built-in worktrees, on all plans. (3) Code review covers GitHub pull requests and GitLab merge requests, on all plans. (4) Codex Security Cloud ships as a plugin for all Codex users. With GPT-6.1 Sol also available in Codex, most of what a Plus subscriber can use right away is in Codex rather than in the chat window.
6. Two new API surfaces: the Decisions API and Agents API computer use
OpenAI describes the Decisions API as focusing Luna's intelligence on a set of user-defined questions, each with a finite set of predefined answers, to make decisions in real time. Developers supply context as text or images, and the answers can classify content, route requests or choose an agent's next action. It launched on September 29 as a limited preview, and OpenAI said at the time that broad availability was planned within days; check OpenAI's documentation for its current status. The difference from a chat call is that you are not asking for prose; you are asking for a judgment that lands inside options you defined, which suits it to acting as a router inside your code. TypeSafe's Jev takes a similar approach (structured answers to questions defined in advance), and we have a separate explainer on it. Computer use in the Agents API is hosted and uses the Codex harness; it is available through the API, and subscribers on the $500 Pro tier and eligible Enterprise customers can also use it in Codex and ChatGPT Work. On the model side, the most direct change for developers is still gpt-6.1-sol, at a standard $2 per million input tokens, $0.10 per million cached input tokens (95% below standard input) and $10 per million output tokens. If your app reuses stable prefixes and repeated context, cache hits make a large difference to cost. A point-by-point comparison of the Decisions API and TypeSafe Jev is at /hub/openai-decisions-api-vs-jev, and Jev itself is covered at /hub/jev-typesafe-ai.
7. By plan: what ChatGPT subscribers get
Breaking OpenAI's wording down by plan, listing only plans OpenAI names explicitly. Plus: GPT-6.1 Sol in ChatGPT Work and Codex (staged launch rollout), Codex Cloud, Codex code review and Codex Security Cloud; dots and the new collaboration features do not list Plus. Pro: the recap lists dots and the new collaboration features (a space for your materials, dynamic pages and collaborative slides) for Pro without saying which Pro tier it means; GPT-6.1 Sol's launch rollout also includes Pro. The $500 Pro tier: OpenAI specifically names the highest usage limits, GPT-6 Astra Ultrafast in ChatGPT Work and Codex, and Agents API computer use in Codex and ChatGPT Work. Business, Business Premium, Enterprise and Edu: GPT-6.1 Sol covers Business, Enterprise and Edu (off by default on Enterprise and Edu until an administrator enables it); dots cover Business Premium and Enterprise; the collaboration features cover Business and Enterprise; Astra Ultrafast covers eligible Enterprise and Edu plans; computer use covers eligible Enterprise customers; sites built with plugins cover Business, Enterprise, Healthcare and Education; privacy intelligence (zero data retention plus the private reasoning preview) is for enterprises. Free: OpenAI says GPT-6.1 Sol does not include Free at launch. The recap labels Codex CLI, code review, plugin expansions, plugin creation and discovery, and event-driven plugin automations via MCP as all plans; for whether anything else reaches the free plan, go by what your ChatGPT account actually shows. If you mostly chat and write, this launch changes little for you directly. If you code in Codex, Plus is in GPT-6.1 Sol's launch rollout and gets Codex Cloud.
8. By role: what developers get in the API
What has launched in the API: (1) gpt-6.1-sol at $2 input, $0.10 cached input and $10 output per million tokens; (2) GPT-6 Astra Ultrafast, which OpenAI says generates up to 6x faster in the API; (3) Agents API computer use, hosted on the Codex harness. Privacy intelligence for enterprises (zero data retention with privacy-safe handling, plus a private reasoning preview) is described in the recap as letting enterprises use frontier intelligence while protecting privacy; it does not say separately how it applies to the API, and private reasoning is still a preview, so it is not on this list. Separately, the Decisions API was in limited preview at launch on September 29, with broad release then planned within days; check OpenAI's documentation for its current status. On model choice, OpenAI says GPT-6.1 Sol brings Sol closer to GPT-6 Astra on hard tasks at one fifth of Astra's standard input and output token prices. Its model documentation positions it for complex coding, computer use and professional work, and says to compare it with Astra on your tasks. For the most demanding scientific research, OpenAI recommends GPT-6 Astra. So before you ship, run your own tasks for cost and quality rather than copying benchmark scores.
9. Practical hurdle one: paying for ChatGPT and the API when your card is declined
For many readers outside the US, the first blocker after the keynote is not a feature but a payment. Paid ChatGPT plans such as Plus and Pro, as well as API top-ups on OpenAI's developer platform, run through cross-border checkouts. One common cause of declines is the card's issuing country, which the acquirer reads from the first digits of the card number (the BIN); a card issued elsewhere can carry a Visa or Mastercard logo and still fail cross-border risk checks. A common approach is a US-issued Visa or Mastercard virtual card, used directly on OpenAI's own pages without handing your account to anyone: (1) sign up for cocodot and top up your wallet; (2) issue a US virtual card in the dashboard and move funds from the wallet onto the card — wallet balance and card balance are separate, and money not moved onto the card cannot be charged; keep enough on the card to cover the full subscription plus a small margin; (3) on the ChatGPT subscription page or the Billing page of OpenAI's developer platform, enter the card number, expiry and CVV, and use a complete address that is genuinely yours (the card is not tied to a personal address, and there is no address on the card page to copy); keep the same address for the same account, and never use one found online, someone else's, or a made-up one. cocodot's US-issued virtual cards have recorded successful charges for ChatGPT subscriptions (the statement descriptor contains OPENAI*CHATGPT SUBSCR) and for OpenAI API top-ups. Issuer and merchant risk rules change, though, and no card can guarantee that every charge goes through. If a charge fails twice in a row, stop and check balance, address and card status rather than retrying repeatedly, so neither the card nor the account gets flagged. One more thing: OpenAI publishes an official list of supported countries and territories (developers.openai.com/api/docs/supported-countries); check it yourself before signing up and using the service.
10. Practical hurdle two: two routes to calling the models
Route one: the official OpenAI API. If you want what was announced — gpt-6.1-sol, GPT-6 Astra Ultrafast, the Decisions API preview or Agents API computer use — cocodot's relay does not offer them, and the most direct route is OpenAI's own platform: pay on the developer platform's Billing page (with the card from the previous section if your local card is declined), create an API key, and call the model ID gpt-6.1-sol following OpenAI's documentation. OpenAI publishes an official list of supported countries and territories (developers.openai.com/api/docs/supported-countries); check it yourself before signing up and using the service. Route two: an OpenAI-compatible relay. If GPT-5.6-generation models are enough for your workload, or you also want Claude, Gemini, DeepSeek, Qwen, GLM, Kimi and others behind one key, cocodot's relay is OpenAI-compatible: change the base_url and key and keep your code. To be clear: cocodot's relay currently offers GPT-5.6 Sol / Terra / Luna, GPT-5.5 and others; it does not offer the GPT-6 family (Astra, Sol, 6.1 Sol, Luna) or the Decisions API. For choosing between the three GPT-5.6 tiers, see our guide at cocodot.co/hub/gpt-56-sol-terra-luna-api; live prices are at cocodot.co/pricing. Below is a minimal request through the relay — confirm model IDs against what /v1/models returns.
curl https://cocodot.co/api/ai/v1/chat/completions \
-H "Authorization: Bearer $COCODOT_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "mog-8-s",
"messages": [
{"role": "user", "content": "Summarize what OpenAI DevDay 2026 means for developers in three sentences."}
]
}'11. A few things to be clear about
First, the DevDay facts on this page come mainly from two official OpenAI posts — the DevDay 2026 recap and the GPT-6.1 Sol announcement, both published on September 29, 2026 — with a few availability details from OpenAI's developer documentation (developers.openai.com) and ChatGPT help documentation (learn.chatgpt.com), as of early October 2026. Where we quote OpenAI's Chinese-language posts, a few feature descriptions are our translation rather than OpenAI's official English names. Prices, plans and availability can change, so check OpenAI's latest pages. Second, monthly prices for ChatGPT plans are whatever OpenAI's pricing page says; this page only cites the $500-per-month tier that the announcement states explicitly. Third, some claims circulating online cannot be traced to official OpenAI pages, so we do not repeat them. Fourth, what cocodot offers here is two things: US-issued virtual cards for subscriptions and API top-ups, and an OpenAI-compatible relay carrying GPT-5.6 and other models. For the GPT-6 family and the Decisions API, go directly to OpenAI.
DevDay 2026 announcements and who can use them (from OpenAI's official posts, checked October 4, 2026)
| Announcement | What it is | ChatGPT / Codex subscribers | Developers (API) |
|---|---|---|---|
| dots | Agents powered by GPT-6 Astra, each with its own cloud computer, working across apps you connect | Pro, Business Premium, Enterprise | No API mentioned in the recap |
| GPT-6.1 Sol | Upgrade to GPT-6 Sol; standard input and output token prices are one fifth of GPT-6 Astra's | Plus, Pro, Business, Enterprise, Edu — in ChatGPT Work and Codex, not yet in chat | Model ID gpt-6.1-sol; per million tokens: $2 input / $0.10 cached input / $10 output |
| Ultrafast | Up to 8x faster token generation in Codex (300 tokens per second), up to 6x in the API | GPT-6 Astra Ultrafast: Pro $500 and eligible Enterprise and Edu plans, in ChatGPT Work and Codex | GPT-6 Astra Ultrafast is live; the recap lists GPT-6.1 Sol Ultrafast as coming soon |
| $500/month Pro tier | New Pro tier with the highest usage limits, including Astra Ultrafast | ChatGPT subscription tier (OpenAI calls it Pro $500) | Not applicable |
| Codex Cloud | Run Codex on your computer, steer it remotely from your phone, or run it in the cloud from any device | The recap names Plus, Pro, Business, Healthcare, Education and Enterprise, and also labels it all plans | Not applicable |
| Codex code review / Codex Security Cloud | Reviews GitHub PRs and GitLab MRs; security delivered as a plugin | Code review: all plans; Security Cloud: all Codex users | Not applicable |
| Decisions API | Applies Luna's intelligence to user-defined questions, each with a finite set of predefined answers, for real-time classification, routing and agent next steps | Not applicable | Limited preview at launch on September 29, with broad release then planned within days; check OpenAI's docs for current status |
| Agents API computer use | Hosted computer use built on the Codex harness | $500 Pro tier and eligible Enterprise customers, in Codex and ChatGPT Work | Available through the API |
| New ChatGPT collaboration features | A space for your materials, dynamic pages and collaborative slides (descriptions translated from OpenAI's Chinese-language recap) | Pro, Business, Enterprise | Not applicable |
| Privacy for enterprises | Zero data retention plus a private reasoning preview | Enterprises (private reasoning is a preview) | The recap does not say separately how it applies to the API |