OpenAI DevDay 2026: Dots personal agent and ChatGPT Space unveiled

OpenAI held DevDay in San Francisco on September 29, 2026, launching Dots, an always-on personal agent powered by GPT-6 Astra, plus ChatGPT Space for human-agent collaboration. It also introduced GPT-6.1 Sol, an UltraFast speed tier, and the Agents API and OpenAI Marketplace.
Key insights
Dots: an always-on personal agent powered by GPT-6 Astra
At DevDay 2026, Sam Altman introduced Dots, a resident online agent with its own cloud computer and browser that can write code and run tests. Dots works with over 4,000 apps connected in ChatGPT, can be reached by text or phone call, and runs on GPT-6 Astra, which Altman called OpenAI's best-aligned model. The agent advances tasks autonomously and seeks approval only at key decision points. Altman said his own Dot reviews overnight messages each morning, and a demo showed a Dot migrating an app off a deprecated API by tracking dependencies, writing code, running tests, and submitting a pull request. Users start with one Dot, with groups of Dots planned.
ChatGPT Space: a shared workspace for humans and agents
ChatGPT Space is a page-based workspace for writing plans, doing research, and generating images and data, with pages and files stored in one place like a cloud drive. Dots can work directly on pages when @-mentioned in comments, and pages can carry standing instructions such as monitoring a Slack channel daily. In the demo, slash commands inserted interactive charts, tables, and runnable prototypes; a Dot named Dotty converted data to a filterable bar chart refreshed hourly and added onboarding data mentioned in Slack private messages to an FAQ. Presentation support in agent-readable formats is planned.
GPT-6.1 Sol and UltraFast target cost and latency
GPT-6.1 Sol delivers capability close to Astra at one-fifth the price, with cached input 95% cheaper than standard input, and Altman said Sol is smarter than Astra in some respects, positioning it as the developer daily workhorse. UltraFast is a new speed tier at eight times speed and six times price, generating 300 tokens per second. Subscriptions were adjusted: the new $500 Pro 500 tier carries 25 times Plus quota with UltraFast access in ChatGPT and Codex, Pro 200 reopened, and a Decisions API was previewed for constrained-choice tasks with sub-second responses. Altman and Tibo also triggered a global usage-quota reset on stage.
Dots already in daily use inside OpenAI
OpenAI product lead Holly demoed Dots managing a product-launch eve: it rescheduled a moved review meeting, triaged overnight tester feedback, and implemented a new homepage design by invoking Codex on her laptop, running the app in an iPhone simulator, and submitting a pull request. Inside OpenAI's Slack, Dots have their own identities, employees route @-mentioned requests to their Dots, and group chats include members' Dots from the start. Engineers' Dots take on bug reports from feedback channels and fix dozens of bugs daily, and parts of the Dots product itself are written by Dots.
Specialist Dots and Microsoft Agent 365 for enterprises
Dots and Space launched to ChatGPT Pro, Business Premium, and Enterprise users, with Dot conversations included in plans and not consuming quota. Enterprise customers get a preview of Specialist Dots: company-configured virtual coworkers for high-volume functions such as accounting, marketing, and legal, given goals and context by the company with outputs reviewed and feedback shared company-wide. OpenAI is also partnering with Microsoft to connect Specialist Dots to Agent 365, Microsoft's tool for managing enterprise agents.
Developer tools: Agents API, Codex cloud, Private Intelligence
OpenAI open-sourced the Codex harness that also powers ChatGPT Work and Dots, and Codex is now fully cloud-based so tasks started on phone continue in browser or desktop. Codex Security Cloud continuously hunts vulnerabilities with verified fixes ready for human review. The new Agents API enters public beta with harness, hosting, memory, multi-agent control, and computer use. OpenAI and AWS launched OpenAI-powered managed agents on Bedrock. A Private Intelligence preview includes zero data retention and private inference for sensitive workloads. The Responses API grew 100 times in a year with 99%+ reliability, 45% faster first token, and 30% faster tool calls.
Models now assist OpenAI's own research
Post-training research lead Tejal described models optimizing their own computer-use harness in a loop, with adopted improvements cutting latency more than two times in production, and models improving monitoring and refusal training that lifted Astra to industry-leading computer-use stress-test results. Internal data shows models completed over a third of day-long research tasks without human intervention by July, up from short 15-minute tasks in January. Astra-class models have helped solve more than 100 open math problems and contribute to work on new antibiotics for resistant infections, ancient languages, renewable energy, and industrial robotics.
Distribution and monetization: ChatGPT as a platform
With ChatGPT at roughly 1.2 billion weekly users, OpenAI launched three distribution mechanisms. Sign in with ChatGPT lets users spend tokens included in their plan inside third-party apps, with 16 launch partners and no model-cost subsidy from developers. Plugin extensions embed editors, dashboards, and whole workspaces natively in ChatGPT and Codex, demoed with a meetings app, Figma, and Adobe Photoshop features, and ChatGPT sites now support plugins and per-user data. The OpenAI Marketplace opened with 30-plus partners including CodeRabbit, Notion, and Vercel, supports enterprise procurement credits, and serves open models through Baseten.