OpenAI DevDay 2026: Every Announcement, From Dots to GPT-6.1 Sol
OpenAI DevDay 2026 was the company's biggest developer event yet. OpenAI's official recap counts more than 20 major announcements, and our own breakdown of the recap comes to 25 separate announcements.
DevDay landed in a busy month for OpenAI's rivals. Anthropic released Claude Sonnet 5.5 the day before, at the same price OpenAI set for GPT-6.1 Sol. Meta and SpaceXAI already had always-on agents on the market. And TypeSafe's Jev, launched two weeks earlier, had developers rethinking how AI should make simple decisions inside software.
OpenAI's announcements pull ChatGPT in two directions. The first is away from waiting for your next prompt, as Dots and team tasks keep working in the background. The second is ChatGPT as a platform, where plugin extensions, Sign in with ChatGPT, and the OpenAI Marketplace point outside products at ChatGPT's 1.2 billion weekly users.
Not everything is live yet. The Decisions API is in limited preview, collaborative slides arrive in the coming weeks, and Private Inference previews this fall.
All 25 OpenAI DevDay 2026 announcements at a glance
Nineteen of the 25 announcements are available now, at least on some plans. The other six are in preview or beta, or are scheduled for the coming weeks.
Table 1 - Every OpenAI DevDay 2026 announcement with plan availability and status, as of October 2026. Source: OpenAI DevDay 2026 recap; Bedrock Managed Agents status per The Next Web.
A whole new way to work with AI
OpenAI opened DevDay 2026 with its biggest bets: an always-on agent and a near-flagship model priced for everyday work, plus faster and more private ways to run them.
1. Dots: always-on agents in ChatGPT
Dots are always-on agents that keep working between conversations, and they were the headline launch of the event. You give a dot goals instead of one-off prompts, and it takes recurring work off your plate.
Each dot runs on GPT-6 Astra and has its own cloud computer. It can connect to more than 4,000 apps and learns from your feedback over time, according to CNBC's live coverage. You can message your dot in ChatGPT on desktop, web, and mobile, or inside Slack and Microsoft Teams. Users set what a dot can do on its own, what needs their approval, and what it must never do, according to Latent Space's AINews summary of the launch.
Dots availability and pricing
Dots are rolling out to Pro and Business Premium users in eligible markets. For Pro, those markets currently exclude the European Economic Area, Switzerland, and the UK, as The Next Web reported. Enterprise, Edu, and Healthcare workspaces can try a beta once an admin turns it on, since it ships off by default.
Your first dot is included in the plan at no extra cost, OpenAI told reporters, according to The Next Web. Chatting with it does not count toward ChatGPT usage limits, but tasks it starts in Codex or ChatGPT Work do. OpenAI also wrote that it envisions teams of dots working together over time.
How Dots compare with Meta Muse and Grok Bot
Dots are OpenAI's entry into a race it did not start. Meta launched its Muse personal agent earlier in September, built around WhatsApp and offered with a free tier. SpaceXAI's Grok Bot gives companies persistent bots for team work, and the open-source OpenClaw had already made always-on assistants popular with developers.
Product writer Peter Yang, who tested Dots early, wrote on X that its biggest advantage is living inside ChatGPT, where his apps and memory were already connected. Reach is the trade-off. At launch, a dot works through ChatGPT, Slack, and Teams, with texting in a limited beta for US Pro users, while Muse is built around WhatsApp.
Note: Emergent's Wingman, launched in April 2026, also works in the background across tools like Gmail, Google Calendar, and Slack, and it pauses for your confirmation before any consequential action. If you need an agent built around one specific job instead, Emergent's AI agent builder creates one from a plain-English description.
2. GPT-6.1 Sol: near-Astra performance at one-fifth the price
GPT-6.1 Sol is an upgrade to GPT-6 Sol that, on OpenAI's own evaluations, nearly matches GPT-6 Astra on agentic coding, computer use, and professional work. It shipped just one week after GPT-6 Sol, and its standard token prices are one-fifth of Astra's.
GPT-6.1 Sol pricing
Input and output prices stay at GPT-6 Sol's levels. The change is cached input, which drops to $0.10 per 1M tokens. That is 95% below the standard input rate and half of GPT-6 Sol's cached price, which matters for agents that reuse the same context across many requests.
Table 2 - GPT-6 lineup standard API pricing, pricing as of October 2026. Source: OpenAI.
GPT-6.1 Sol benchmarks
OpenAI's launch post backs the pricing pitch with benchmark results. These are OpenAI-reported evaluations, not independent head-to-head tests, and OpenAI notes that results can differ from production ChatGPT because of system prompts, tools, and reasoning settings.
Table 3 - GPT-6.1 Sol benchmark results as reported by OpenAI (vendor-reported, not independently verified). Source: OpenAI.
How GPT-6.1 Sol compares with Claude Sonnet 5.5
GPT-6.1 Sol also arrived in a crowded week. Claude Sonnet 5.5 launched on September 28 at the same $2 input and $10 output list price. It was the second week running that the two companies shipped rival models within a day of each other, after GPT-6 Sol and Claude Opus 5.5 the week before.
With list prices tied, the differences sit in the details. GPT-6.1 Sol charges half of Sonnet 5.5's $0.20 cached input rate. But it bills a whole request at higher rates once the prompt passes 272,000 tokens, while Sonnet 5.5 keeps one rate across its 1M-token window.
Independent testing points both ways. On the Artificial Analysis Intelligence Index, Sonnet 5.5 scores 56 to GPT-6.1 Sol's 52, as reported by DataCamp and Codersera. Sol costs far less per completed task, about $0.72 against $7.60, because Sonnet 5.5 generates many more output tokens.
GPT-6.1 Sol availability
The model carries a 1,050,000-token context window and up to 128,000 output tokens, per OpenAI's model documentation. Developers call it as gpt-6.1-sol in the API. In ChatGPT, it is available to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, though not yet in regular Chat. For a breakdown of the previous version's costs, see our GPT-6 Sol pricing guide.
GPT-6.1 Sol is already live on Emergent, so you can build with it through the Universal LLM Key without setting up a separate OpenAI account.
3. Ultrafast: a premium speed tier
Ultrafast is a new premium speed tier for workloads where response time matters most. It generates tokens up to 8x faster in Codex, reaching 300 tokens per second, and up to 6x faster in the API.
GPT-6 Astra Ultrafast is available now in the API and in ChatGPT Work and Codex on the Pro 500 and Enterprise plans. GPT-6.1 Sol Ultrafast is due in the coming days.
Speed has become its own competitive front. OpenAI first previewed Ultrafast in August for GPT-5.6 Sol, running on Cerebras chips at up to 750 tokens per second for a limited group of customers, as Cerebras announced at the time. Anthropic offers its own Fast mode for Claude Opus models.
Ultrafast is separate from OpenAI's existing Fast mode, which GPT-6.1 Sol's model page lists at 2x Standard rates. According to VentureBeat, OpenAI prices Ultrafast at six times Standard rates. That would put GPT-6 Astra Ultrafast at $60 input and $300 output per 1M tokens.
Speed compounds in agentic work. An agent that iterates on code over dozens of steps spends most of its time generating tokens, so a faster tier shortens the whole loop, not just one reply.
4. Private Intelligence
Private Intelligence is OpenAI's new package for businesses that need stronger data protection. Its first piece, Zero Data Retention with Private Safety Processing, runs automated safety reviews without OpenAI staff seeing the underlying content. A preview of Private Inference, which pairs confidential computing with verifiable controls, follows this fall.
New tools for developers in Codex and the API
Developers got seven launches, split between Codex upgrades and new API building blocks for agents.
5. Codex in the cloud
Codex now runs wherever you work: on your computer, remotely from your phone, or in the cloud from any device. Reusable development environments help tasks start quickly and give a team one shared setup with approved settings and permissions. It is available on Plus, Pro, Business, Healthcare, Education, and Enterprise.
6. A refreshed Codex CLI
The Codex CLI now lets you start and steer tasks by voice. A new /agents view makes it easier to delegate work and track several tasks at once, and OpenAI cleaned up prompt editing, session resuming, worktrees, and the terminal UI. It is available on all plans.
7. Code Review in the ChatGPT desktop app
A new code review experience in the ChatGPT desktop app shows summaries and diffs across projects. You can ask Codex about potential issues before leaving feedback on GitHub pull requests or GitLab merge requests. Automatic reviews let Codex take a first pass in the cloud while you are away, on all plans.
8. Codex Security Cloud
Codex Security Cloud scans entire GitHub repositories on demand or on a schedule, then keeps checking new commits. Codex investigates findings, removes duplicates, and prepares fixes in the cloud, even with your laptop closed. It includes access to Daybreak Blue models without a separate application and is available to Pro, Business, Enterprise, and Edu users.
Security Cloud turns Daybreak, OpenAI's controlled-access cybersecurity program, into a product. OpenAI launched Daybreak in May and split it into Blue and Red tiers in August, with Blue as the starting point for defensive work. Until now, defenders had to apply to get in.
9. Decisions API: fast, constrained answers from Luna
The Decisions API turns a model into a fast, constrained judge. You define a set of questions, each with a fixed list of allowed answers, and the API focuses OpenAI's Luna model on those questions alone.
How the Decisions API works
You send context as text or images and get back one of the answers you defined. OpenAI pitches three uses: classifying content, routing requests, and choosing an agent's next action.
The constraint is the point. Free-form model output has to be parsed and can drift from the format you expected. A fixed answer set gives your app something it can branch on directly, such as sending a support ticket to billing or to technical support. That makes it a natural fit for the small, frequent decisions inside a larger agent workflow.
How the Decisions API fits the market
OpenAI is not alone here. TypeSafe AI launched Jev two weeks earlier, on September 15, as a "decision model" that returns typed answers with confidence scores instead of text. Backed by a $40 million seed round led by DCVC, Jev reached 13% of paid teams on Vercel's AI Gateway within 24 hours, The Next Web reported. Some developer coverage, including daily.dev and Latent Space's AINews, framed the Decisions API as a response to Jev.
The two take different routes. Jev is a separate model trained only for decisions, while OpenAI's API runs on Luna, a general model it already serves at scale. For teams already on OpenAI, that means decision-style outputs without adding another provider.
The API is in limited preview, with a broad release planned in the coming days. OpenAI has not published separate pricing for it yet.
10. Agents API with computer use
The Agents API now supports computer use, so developers can build agents that operate software through its interface to finish a task. That matters because many business tools have no API, and computer use lets an agent work through the same screens a person would.
What the Agents API adds
The update also brings Codex's multi-agent capabilities, tool search, tool calling, and context compaction into your own application. Context compaction keeps long-running tasks going without exhausting the context window. OpenAI runs the underlying infrastructure, so teams do not host the agent loop themselves.
Computer use is available through the API, as covered in OpenAI's computer use guide. It is also available in Codex and ChatGPT Work on the Pro 500 and Enterprise plans. The timing pairs well with GPT-6.1 Sol, whose biggest reported gains include the OSWorld 2.0 computer-use benchmark.
How the Agents API compares with Anthropic's computer use
Computer use is not new to the industry. Anthropic has offered it for Claude since October 2024. OpenAI's angle is hosting: the Agents API, in public beta since earlier in September, runs agents in a managed sandbox with an OpenAI-hosted browser. Developers get the Codex harness without building and running that infrastructure themselves.
11. Bedrock Managed Agents, powered by OpenAI
OpenAI worked with Amazon on Bedrock Managed Agents, which takes the core of the Agents API and adapts it to run natively in AWS. Teams can now build OpenAI agents that run entirely inside AWS and connect to their existing AWS resources.
The deal matters because Bedrock has long been one of Anthropic's main distribution channels, and Amazon is a major Anthropic investor. For AWS customers, it removes a common blocker to trying OpenAI agents: all inference runs on Bedrock and customer data stays inside AWS, per The Next Web. The service is in limited preview.
ChatGPT becomes a platform for developers
The biggest structural shift at OpenAI DevDay 2026 is that ChatGPT is now a surface other companies build on. OpenAI is opening the platform it uses to build ChatGPT's own features, so developers can launch native experiences directly to ChatGPT's 1.2 billion weekly users.
Plugins now get the same building blocks as OpenAI's first-party features: a place in the sidebar, interactive panels, file viewers, and automations triggered by events. Users stay in control, choosing which plugins to use and approving the access each one receives.
OpenAI has tried to build an app layer inside ChatGPT before, from the 2023 plugins beta to the GPT Store in 2024 and the Apps SDK at DevDay 2025. This version goes further. Outside products now get a permanent place in ChatGPT's interface and can trigger work on their own, instead of waiting inside a chat to be called.
Four launches make up this shift.
12. Plugin extensions
Plugin extensions give your plugin a home in the ChatGPT sidebar, along with interactive panels where people work alongside the conversation. You can also build viewers for the file types your product supports. OpenAI's extensions documentation covers the build process, and the feature is available on all plans.
13. Improved plugin creation, submission, and discovery
A new Plugin Creator helps you build a plugin, and a redesigned submission flow gives clearer feedback during review. Better ranking and recommendations help people find relevant plugins, both in the directory and inside conversations. This applies to all plans.
14. Sites can now host plugins
You can now add supported ChatGPT plugins to the Sites you build in ChatGPT. Teammates use the same app with their own connected data and permissions, and automations are easier to add so a Site's shared information stays current. It is available on Business, Enterprise, Healthcare, and Edu plans.
15. MCP events for plugin automations
ChatGPT now supports the proposed MCP Events specification, so a plugin can start an automation when something happens in a connected app. For example, ChatGPT can watch a project board for new tasks, read the linked documents, and draft a plan while you are away. It is available on all plans.
Improving how people and AI work together
Seven launches turn ChatGPT into a shared workspace where teammates and agents edit the same files. Latent Space's AINews newsletter read Space and Pages as OpenAI's answer to Notion and the office suite, a notable move given that Notion is also a Sign in with ChatGPT launch partner.
16. ChatGPT Space
ChatGPT Space is a dedicated home where teammates, ChatGPT, and your dot build on shared knowledge. ChatGPT keeps the space organized based on instructions you give it, so work is easy to find and pick up again. It is available on Pro, Business, and Enterprise via desktop and web, with mobile limited to finding, reading, and sharing pages for now.
17. Pages
Pages are a new document type built for people and agents to work on together. Anything you can do in ChatGPT works on a page, from research and writing to charts and images. You create a page in a conversation and invite your team to contribute, on Pro, Business, and Enterprise plans.
18. Collaborative slides
Collaborative slides arrive in the coming weeks for Pro, Business, and Enterprise plans. Teammates and agents will be able to edit a deck at the same time, starting from a conversation or your own template. You can present inside ChatGPT or export to PowerPoint or Google Slides with formatting intact.
19. Teams and shared tasks
You can now group colleagues into a team in ChatGPT to share pages, slides, plugins, and spreadsheets. Recurring work, like a weekly project update, can be handed to team tasks that run on a schedule or fire on events such as a new email or Slack message. It is available on Business and Enterprise plans.
20. @ChatGPT in Slack and Microsoft Teams
You can now mention @ChatGPT in a Slack or Microsoft Teams channel, thread, or DM to get help with team work. It uses tools your admin connected or your own tools under your permissions. Teammates can add context in the same thread without needing their own ChatGPT license, on Business and Enterprise plans.
21. Meetings plugin
The Meetings plugin takes notes and saves personalized summaries with action items to ChatGPT Space. You can keep notes private or share them, then ask ChatGPT to update a project plan or draft a follow-up. Audio is deleted once notes are ready, and the beta runs in the macOS desktop app for Pro and Business users, with Enterprise coming soon.
22. Shareable profiles
Shareable profiles let people showcase the Sites and plugins they have built so others can find and reuse them. Teammates can also discover shared skills, though skill sharing stays within a workspace. Profiles are available on Free, Go, Plus, Pro, Business, and Enterprise plans.
Do more with your ChatGPT subscription
The last three launches extend what a ChatGPT plan pays for, both inside ChatGPT and in other companies' tools.
23. Sign in with ChatGPT
Sign in with ChatGPT lets you spend your ChatGPT plan inside other companies' tools. You can use your allowance across 16 partners, including Cognition's Devin, Notion, Vercel, T3, OpenClaw, and Dactyl, and cap how much each one can use. Eligible OpenAI usage in those tools counts toward your plan limits.
What changed at DevDay
The login itself is not new. It launched in beta in July with partners including Airtable, GitLab, HubSpot, Notion, Supabase, and Vercel. What DevDay added is plan usage, which is open to open-source projects, locally running apps, and selected partners, with other commercial tools on a waitlist.
The feature works on two levels. Identity lets you sign in to supported tools with your ChatGPT account, which means faster plugin connections and fewer passwords, and it is available globally. Plan usage, where a partner tool draws on your subscription, is limited to Plus and Pro users in participating tools.
Why Sign in with ChatGPT matters for developers
For developers, this changes who pays for AI usage. A product that adopts it can let users draw on the ChatGPT subscription they already pay for, rather than passing API costs through in its own pricing. It also keeps users inside OpenAI's ecosystem even when they work in another company's product. Once the allowance runs out, usage stops rather than switching to the app's own billing. OpenAI's help article explains how users manage connected tools.
24. A new Pro tier: Pro 500
Pro 500 is a new ChatGPT plan with OpenAI's highest usage allowance, set at 25 times the Plus allowance, and it includes access to Ultrafast. Neowin reports that it costs $500 a month and sits above the existing $100 and $200 Pro tiers. It also reports that OpenAI lowered the usage allowance on the $200 tier at the same time.
25. OpenAI Marketplace
The OpenAI Marketplace lets eligible enterprise customers apply part of their existing OpenAI commitment toward approved partner software. The first 32 partners span creative work, customer experience, legal, and cybersecurity, with names such as Figma, Salesforce, Harvey, and CrowdStrike. Enterprise customers can register interest now.
The model will be familiar from cloud marketplaces, where committed spend can go toward third-party software. For OpenAI, it gives enterprises a reason to sign larger commitments, since that money is no longer locked to OpenAI's own products.
What to act on first after OpenAI DevDay 2026
OpenAI DevDay 2026 landed in a crowded market. GPT-6.1 Sol arrived a day after Claude Sonnet 5.5 at the same price, and Dots joined a race Meta and SpaceXAI had already entered. The Decisions API followed TypeSafe's Jev by two weeks.
Underneath the individual launches sits one bet. OpenAI wants ChatGPT to be the place where agents work in the background and outside products reach 1.2 billion weekly users.
For most builders, GPT-6.1 Sol is the launch to act on first. It brings near-Astra results on coding and computer use at Sol's $2 input and $10 output price, and it is available today. Plan the rest around rollout dates: the Decisions API and GPT-6.1 Sol Ultrafast are days away, while collaborative slides and Private Inference come later.
GPT-6.1 Sol is live on Emergent now through the Universal LLM Key. Describe the app you want, and Emergent's multi-agent architecture builds and deploys it with a real backend.
on Emergent today
