OpenAI DevDay 2026: What Was Announced, What the Demos Showed, and What Comes Next

OpenAI announced more than 20 updates at DevDay on 29 September 2026. The easiest way to understand them is through three questions:

1. Can AI keep working on a goal after you leave the chat? That is what Dots are for.
2. Can people and agents work on the same thing? That is what ChatGPT Space is for.
3. Can developers build those systems without managing every step? That is what the expanded Agents API, Codex and GPT-6.1 Sol are for.

This is a plain-English guide to what was announced, what the demos showed, who spoke, what you can use today and what is still coming. It also covers what happened the day before, which changes how you should read the rest. Availability notes describe launch day, 29 September 2026. Check the linked OpenAI pages for later changes.

1. Everything OpenAI announced, in one table

Three distinctions matter more than the list. A Dot is an agent. Space is a place for shared work. GPT-6.1 Sol is a model. Getting one does not mean you get the others. Each has its own plans, regions and limits. Confirmed

2. Which announcement matters to you?

Most people only need two or three of the 20+ announcements. Start with the row that describes you.

3. The day before: GPT-6.1 Astra was held back

On Monday 28 September, the day before DevDay, OpenAI confirmed it would not release GPT-6.1 Astra, the next version of its top model. Internal tests showed it did not meet OpenAI's safety bar. Reported

Saachi Jain, OpenAI's head of safety systems, said the model "didn't quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it's done." Reports say testing found higher levels of deception and a tendency to push ahead on tasks without the user's permission. Reported (Al Jazeera)

This matters for reading DevDay. The headline launch the next day was Dots, an agent built to work without you watching. It runs on GPT-6 Astra, the model already on sale, not the one that was held back. It also came weeks after an OpenAI agent accessed files on an Australian government portal without being asked, covered in AI & SEO Weekly #3. Watch how OpenAI's approval controls hold up in practice, not just how the demos look. Speculation

4. The keynote: who spoke and what was shown

OpenAI's event schedule named CEO Sam Altman for the opening keynote. Confirmed Altman introduced Dots and showed them working on website designs, slide decks and calendar management. OpenAI's Holly Li, from the product team, demonstrated talking to a named Dot. Reported (Barron's)

The live voice demo lagged. Observed (Axios) A stage demo shows how a product is meant to work. It is not a test of how reliably it works on an ordinary account.

Three moments from the stage are worth knowing:

• "We now have an AI research intern." Altman said OpenAI set this goal a year ago and reached it a few weeks ago. The company says its models now finish more than a third of day-long research tasks with no human help. No paper or outside evaluation has been published. Treat it as OpenAI's claim for now. Reported (The Next Web)
• No IPO rush. Altman said he has no set timeline for going public and does not want the distraction while safety and alignment questions remain open. CFO Sarah Friar said OpenAI will go public "when the time is right for our business" and will "pace the frontier" when needed. Reported (CNBC)
• Shared context was the real demo. The point of Space and Dots is that the assistant works from your team's current plan, not a prompt you copied into a new chat. Confirmed

OpenAI's written recap lists the launches but is not a transcript of the stage. Speaker and demo details above come from named press coverage. Feature and rollout details below come from OpenAI's own pages.

5. What are OpenAI Dots, and how are they different from ChatGPT?

A Dot is an agent that keeps working toward a goal, uses your connected apps, and reports back after your conversation ends. Confirmed (OpenAI)

• Brain: runs on GPT-6 Astra.
• Hands: its own cloud computer and browser, plus more than 4,000 apps through OpenAI's plugins.
• Where you talk to it: ChatGPT, Slack, Microsoft Teams, or by voice call. Texting is coming later.
• How many: one main Dot per person at launch. More Dots, and specialist Dots for companies, come later or in pilots.

OpenAI's own examples include revising launch materials after the product changes, preparing a tested software fix, and updating an analysis when new evidence arrives. Those show the intended use. They do not prove every customer can reproduce them without setup, access and review. Confirmed

Who can get a Dot?

• OpenAI says the first Dot is included in eligible Pro and Business Premium plans, and deeper work has a usage allowance. Tasks started in Codex or ChatGPT Work count against those products' limits. Confirmed
• CNBC reports Dots come with the $200 Pro plan and Business Premium, not the $100 Pro plan. Check your own plan page before upgrading. Reported
• Pro users in the EU, the UK and Switzerland do not get Dots at launch, and OpenAI has not said when they will. Business Premium is available in all supported regions. Reported (The Next Web)
• Enterprise, Edu and Healthcare get a beta that is off by default until an admin turns it on. Confirmed
• The price of extra Dots has not been announced. Reported

6. Safety and control: what a Dot can do without asking

Read this before you connect a work account. According to OpenAI: Confirmed

• You choose which apps a Dot can connect to, and what permissions it has in each.
• Custom Rules can allow an action, block it, or require your approval first.
• Auto-review checks actions that could affect accounts or share information. Some actions need your approval.
• Sensitive actions stay with you, such as changing a password.
• Background research is read-only. "Proactive research" uses restricted tools that cannot change anything.
• You can inspect or pause a Dot's work at any time.
• OpenAI says plainly that Dots can make mistakes.

My advice: start with read-only connections, turn on approval for anything that sends, publishes, pays or deletes, and review the activity log for the first two weeks. My guide on running a computer-use agent safely covers the setup in detail. Speculation

7. What is ChatGPT Space, and what are Pages?

Space is a shared home for work. Pages are editable documents inside it. A team can bring files and plans together, edit a Page, leave comments, and ask ChatGPT, Codex or a Dot to update or check something. Confirmed (OpenAI)

• Space replaces Library for Pro, Business and Enterprise users.
• Web and desktop: create and edit Pages.
• Mobile: find, read and share Pages. Mobile editing comes later.

The difference from a folder of AI chats is that the work itself becomes the shared thing. An SEO team could keep a content brief, source list, change log and open questions on one Page. A teammate fixes a claim, an agent checks a linked document, and an editor approves the final wording. That is my suggested workflow, not a DevDay demo, and it does not make anything automatically accurate. Speculation

Also announced for teams: Confirmed

• Collaborative slides in the coming weeks, with live editing and export to PowerPoint or Google Slides.
• Team tasks for Business and Enterprise, which run on a schedule or start when something happens in a connected app.
• A Meetings plugin that writes notes and action items into Space. The first beta is for Pro and Business on the macOS desktop app, with Enterprise later.

8. What is GPT-6.1 Sol, and can you use it in ChatGPT?

GPT-6.1 Sol is an upgraded model for coding, computer use, document-heavy work and agents. OpenAI says it gets close to GPT-6 Astra on selected tests at one-fifth of Astra's token prices. That is OpenAI's comparison, not an independent finding that it matches Astra on every task. Confirmed (OpenAI)

The list price is the same as GPT-6 Sol from a week earlier. The change is the model behind it and the very cheap cached input. If your app reuses a large, stable context, such as a codebase, a product catalogue or a style guide, the $0.10 cached rate can matter more than anything else here. Your real cost still depends on cache hits, tool use, retries and review time. Confirmed

The benchmark claims, in plain words

These are OpenAI's numbers, as reported by press covering the launch. Nobody has independently checked them yet. Reported (The Next Web, Vellum)

• Coding (DeepSWE v1.1): about 74.8%, matching GPT-6 Astra's best score at roughly 80% lower cost per task.
• Automation (AutomationBench): 2.2 points ahead of Claude Opus 5.5, at about a third of the cost.
• Using a computer (OSWorld 2.0): within 2.1 points of Astra, at about a seventh of the cost per task.

One caution from OpenAI's own fine print: its factuality test uses hard conversations where a user had already flagged an error, and OpenAI says those prompts are not typical of normal use. The sensible test is your own: run real tasks and measure finished work, error rate, speed and how long a person spends reviewing. Confirmed For how the Sol family compares with Claude, see Claude Opus 5.5 vs GPT-6 Sol.

9. What changed for developers and Codex users

The developer announcements fit together as a stack:

1. Codex in the cloud. Start or continue a coding task from your computer, phone or any device, with reusable cloud environments for a shared setup. The command-line tool gains voice steering and an /agents view for following delegated work. Confirmed
2. Agents API computer use. Your app can let an agent use software through a browser hosted by OpenAI. Your app handles website approvals and sign-ins. OpenAI also lists multi-agent features, tool search, tool calling and context compaction. The Agents API itself entered public beta earlier in September; DevDay expanded it. Confirmed (API changelog)
3. Decisions API. You give it context and a fixed list of allowed answers, and it returns one, for example to classify a request, route a ticket or pick an agent's next step. Confirmed It answers in a fraction of a second, uses GPT-6 Luna under the hood and returns a confidence score. It is in limited preview with no pricing yet. Reported (Simon Willison, The New Stack via Techmeme) Use it for bounded, low-stakes choices, not for high-stakes judgement nobody reviews.
4. Plugin extensions and event-driven work. Plugins can now be full applications inside ChatGPT, with sidebar homes and panels. OpenAI also backs a proposed MCP Events specification, so an event in a connected app can start an automation. It is only a proposal, so check its status before building on it. Confirmed

For code quality: a desktop code-review experience, automatic cloud reviews, and Codex Security Cloud, which scans repositories, investigates findings and prepares fixes. A prepared fix is not a verified fix. Tests, review and a named owner still matter. Confirmed (The Decoder)

Speed and privacy:

• Ultrafast is a premium speed tier: up to 8× faster in Codex, around 300 tokens a second, and up to 6× faster in the API. Astra Ultrafast is available now on the API and top plans. Sol Ultrafast comes later. Reported
• Private Intelligence adds a Zero Data Retention option for safety processing now. A separate Private Inference preview is due in the autumn. They have different availability and controls. Confirmed

If you are moving from GPT-6 Astra, my GPT-6 Astra API guide has the rate-limit and sandboxing notes that still apply.

10. Plans, prices and where you can get it

DevDay also changed what the paid plans include. This is the part most recaps skip. Reported (Engadget, The Next Web)

If you pay for Pro 200 mainly for Codex, check your usage before 30 October. The same plan will cover half as much Work and Codex use from that date.

Sign in with ChatGPT lets Plus and Pro users spend their plan's allowance inside 16 partner tools at launch, including Devin, Notion and Vercel. You can set a weekly cap for each app. Reported

The official recap also lists ChatGPT in Slack and Microsoft Teams, easier plugin building and discovery, ChatGPT Sites hosting plugins, shareable profiles, an enterprise Marketplace, and Bedrock Managed Agents built with Amazon for AWS. Prices and regions vary, so check the live pricing page. Confirmed

11. What is available now, and what is coming next?

These are launch-day snapshots, not permanent rules. The recap, the Dots page and the model announcement are the places to check for changes.

12. What DevDay 2026 means for content, search and marketing teams

My reading: OpenAI is making continuity, connected tools and reviewable actions the centre of everyday AI work. That matters more to a marketer than any single benchmark. A workflow can now be built around a living source of truth: what changed, which evidence supports it, who approved it, and what an agent may update. Speculation

Here is a narrow experiment I would run before connecting an agent to publishing tools:

That is my recommendation, not an OpenAI guarantee. For GEO and AEO, the opportunity is clearer, better-sourced facts that a reader or answer engine can lift. There is no evidence that using Space, Dots or a new model improves Google rankings or earns AI citations by itself. The editorial quality and the source trail still do the work. See how to get cited by AI search and what new models change for SEO.

13. Frequently asked questions

What was announced at OpenAI DevDay 2026?

OpenAI announced more than 20 updates on 29 September 2026. The main ones were Dots, which are always-on agents with their own cloud computer. ChatGPT Space and Pages add shared workspaces for people and agents. GPT-6.1 Sol is a model near GPT-6 Astra's level at a fifth of the price. Codex gained cloud work and security scanning, the Agents API gained computer use, and a new Decisions API picks answers from a fixed list. OpenAI also added Ultrafast speed, a $500 Pro 500 plan and Sign in with ChatGPT.

When and where was OpenAI DevDay 2026?

It took place on 29 September 2026 at Fort Mason in San Francisco, with about 2,500 developers attending. Sam Altman gave the opening keynote, followed by technical sessions and demos.

What are OpenAI Dots?

Dots are always-on AI agents in ChatGPT. Each Dot runs on GPT-6 Astra, has its own cloud computer and browser, and can reach more than 4,000 apps through plugins. You give it a goal, and it keeps working after the chat ends, then reports back or asks for approval. You can reach it in ChatGPT, Slack, Microsoft Teams or by voice call.

How much do Dots cost?

One Dot is included in eligible Pro and Business Premium plans. CNBC reports that means the $200 Pro plan, not the $100 one. The price of extra Dots has not been announced. Pro users in the EU, the UK and Switzerland do not get Dots at launch.

Is a Dot the same as a custom GPT?

No. A custom GPT answers inside a chat. A Dot is a persistent agent with its own cloud computer, connected apps, ongoing goals, permissions and action review. It keeps working after you leave.

Can I use GPT-6.1 Sol in regular ChatGPT?

Not at launch. OpenAI made GPT-6.1 Sol available in the API as gpt-6.1-sol, and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. It said the model was not yet in regular ChatGPT chat.

How much does GPT-6.1 Sol cost?

GPT-6.1 Sol costs $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens in the API. OpenAI says that is a fifth of GPT-6 Astra's standard prices.

Did OpenAI release GPT-6.1 Astra at DevDay?

No. On 28 September 2026, the day before DevDay, OpenAI said it would not release GPT-6.1 Astra because internal testing showed it did not stay within scope and authorisation well enough. The model released at DevDay was GPT-6.1 Sol. GPT-6 Astra, released earlier in September, is still OpenAI's top available model.

What is the Pro 500 plan?

Pro 500 is a new ChatGPT plan at $500 a month, announced at DevDay 2026. It has OpenAI's highest usage allowance and is the only Pro plan with Ultrafast speed. From 30 October 2026, the $200 Pro plan's allowance for ChatGPT Work and Codex falls from 20 times to 10 times the Plus allowance.

Does Space replace Library?

Yes, for Pro, Business and Enterprise users. OpenAI says Library is now Space, with Pages and shared work added. At launch, Pages can be edited on web and desktop, while mobile can find, read and share them.

Can a Dot act without asking me every time?

Within limits you set, yes. Custom Rules can allow, block or require approval for actions. Auto-review checks risky actions, some actions always need your approval, and sensitive ones such as password changes stay with you. Background research is read-only, and you can pause a Dot at any time.

Work with me

Get your site ready for agents that browse, compare and buy

I'm Jayant Solanki, an SEO, GEO and automation strategist working with eCommerce, local-service and global brands. Always-on agents mean more visits from software acting for a person. I build sites those agents can read, trust and finish tasks on.

An engagement typically covers:

• A retrieval audit: crawler access, rendering and indexation on revenue pages
• Content restructured so a model can lift a specific claim cleanly
• Removing agent friction from forms, checkouts and gated flows
• Safe, human-approved AI workflows for content and product updates
• Measurement that separates agent traffic from human traffic

Book a technical review → Is your site agent-ready?

Ranked #1 for "metal buildings" · +30% YoY organic traffic · Evidence-labelled research

Sources

• OpenAI, DevDay 2026 Recap: the official announcement list and rollout notes
• OpenAI, Introducing dots: agent design, examples, permissions, safety and access
• OpenAI, GPT-6.1 Sol: model claims, benchmark conditions, pricing and availability
• OpenAI, ChatGPT Space: Pages, collaboration and device support
• OpenAI, API changelog and DevDay event page
• CNBC, OpenAI DevDay recap: Dots agents, Altman and Friar comment on IPO
• Axios, The 5 biggest announcements and the Dots live demo
• Barron's, DevDay keynote report
• The Next Web, Dots, GPT-6.1 Sol, Pro 500 and the Pro 200 cut and the "AI research intern" claim
• Engadget, OpenAI adds a $500 Pro subscription
• Al Jazeera, OpenAI cancels release of GPT-6.1 Astra over safety concerns
• The Decoder, Codex security scans, the Decisions API and Ultrafast
• Simon Willison, OpenAI DevDay 2026 live blog
• Vellum, GPT-6.1 Sol benchmarks explained

Editorial note: this guide is based on OpenAI's published launch materials and named press coverage of the keynote. It does not claim hands-on testing of the new products. Rollout and pricing details reflect launch day and will be refreshed as OpenAI updates them.

Jayant Solanki

AI-Ready SEO, GEO & AIO strategist based in Indore, India, working with eCommerce, local-service and global brands across India, the UAE and the US. This page is maintained and will be updated as OpenAI's DevDay features roll out more widely.

Google Analytics Certified (2026–2027) · verify

More about Jayant →