OpenAI’s September 29 DevDay covered much more than a model release. Always-on agents, shared documents, cloud coding, new APIs and plugins all pointed in the same direction: more work happening inside an AI workspace, with people and agents returning to it over time.
Here is the complete product announcement roundup, grouped by what changed. Availability reflects the launch announcements and documentation checked on September 30. Where detailed documentation narrows a broad recap claim, the detail appears below.
Always-on agents
Dots
Dots are persistent agents powered by GPT-6 Astra, with their own cloud computer and access to connected apps. You can inspect their work and set rules for independent actions, approvals and blocked actions. Background proactive research uses read-only tools. Rollout begins on Pro and Business Premium in eligible markets; Enterprise, Edu and Healthcare get an admin-enabled beta. Source.
Specialist dots and Microsoft Agent 365
OpenAI previewed specialist dots for defined organizational responsibilities, with their own identities and deeper enterprise integrations. A Microsoft Agent 365 integration is planned. Teams of dots and texting are future plans, rather than features everyone can use today. Source.
Models, speed and privacy
GPT-6.1 Sol
Sol gets an upgrade for coding, computer use and professional tasks. OpenAI says it approaches Astra on selected evaluations at one-fifth of Astra’s standard token prices. API pricing is $2 per million input tokens, $0.10 for cached input and $10 for output. It launched in the API, ChatGPT Work and Codex for eligible users, but not ordinary Chat. Source.
Ultrafast
A paid speed tier offers faster token generation. OpenAI reports up to eight times faster generation in Codex and six times in the API. Astra Ultrafast is available; Sol Ultrafast is coming soon. Those are generation-speed claims, not guaranteed end-to-end task savings. Source.
Private Intelligence
Zero Data Retention with Private Safety Processing supports automated safety review while preventing OpenAI personnel from reading protected content. Customer-controlled storage and access controls are part of that design. Separately, OpenAI announced a Private Inference preview for this fall, combining confidential computing with verifiable controls. Source. Private Inference announcement.
Codex keeps working in the cloud
Codex Cloud
Coding tasks can continue while your computer sleeps. Reusable cloud environments give projects a shared setup, and developers can review and continue work from web, mobile or desktop. The practical change is being able to leave a task running and return from another device. Source.
A refreshed CLI
The terminal experience adds voice steering and an /agents view for tracking delegated work. OpenAI also announced improvements to prompt editing, session resumption and worktree workflows. Source.
Code Review
The review interface brings summaries, changed files, diffs and questions about a change together. Codex can perform automatic reviews in the cloud, while people inspect findings and decide what feedback to send. GitHub reviews are generally available; GitLab support is in preview. Source.
Codex Security Cloud
The security workflow scans connected GitHub repositories in the cloud, investigates findings and prepares fixes. Teams can monitor new commits as well as run repository scans. The cloud service is distinct from the local Codex Security plugin. Source.
More ways to build agents
Agents API with computer use
The Agents API was introduced earlier in September. DevDay adds computer use, so applications can run agents that interact with software interfaces. The broader managed runtime includes context management, tool calling and multi-agent work. This is an expansion of the API, rather than its first announcement. Source. Computer-use documentation.
Decisions API
Luna can answer developer-defined questions with finite, predefined answers. Those outputs can classify content, route requests or choose an agent’s next action. It is in limited preview, with broader release planned in the coming days. Source.
Amazon Bedrock Managed Agents
AWS and OpenAI introduced a preview that combines OpenAI models, the Codex harness and Amazon Bedrock AgentCore. Agents run inside AWS, using its identities, permissions and governance controls. This gives AWS teams another deployment option. Source.
ChatGPT opens more of its interface
Plugin extensions
Developers can give plugins sidebar homes, panels beside conversations, and custom file viewers and editors. Extensions also support settings and richer forms. The documentation says web extensions for Free and Go are coming soon, so the recap’s broad plan availability should not be read as every extension already working everywhere. Source.
Plugin creation, submission and discovery
Plugin Creator, clearer submission feedback, and improved ranking and recommendations are intended to make plugins easier to build and find. Users still choose the plugins they install and approve their access. Source.
Plugins inside Sites
Supported plugins can be added to Sites, letting teammates use the same app with their own connected data and permissions. OpenAI also announced easier automation management for keeping shared information current. Source.
MCP events
Plugins can deliver events from connected applications to start work. A project-board update, for example, can trigger an automation instead of waiting for a person to ask again. The documentation describes subscriptions, filters and authenticated event delivery. Source.
Shared work, documents and conversations
ChatGPT Space and Pages
Space is a shared home for files, project context and work with teammates and agents. Pages are collaborative documents where people can write, research and create visuals with AI. Pro, Business and Enterprise users can create and edit on web and desktop; mobile supports finding, reading and sharing pages at launch. Source.
Collaborative slides
OpenAI announced slides that teammates and agents can edit together, with comments and exports to PowerPoint or Google Slides. The announced timing is the coming weeks, rather than general availability on DevDay. Source.
Teams and Team Tasks
Teams can organize shared work and delegate recurring tasks, either on a schedule or in response to an event. Tasks run in the cloud using the team’s service account and configured app connections. Those connections determine which external accounts the work uses. Source.
ChatGPT in Slack and Microsoft Teams
People can mention @ChatGPT in channels or threads, add context together, and ask it to use connected tools. An individual ChatGPT license is not required for each participant in an enabled channel. Access depends on the organization’s setup and permissions. Source.
Meetings plugin
Meetings creates personalized summaries and action items in Space from online or in-person conversations. It is in beta on macOS for Pro and Business, with a limited Enterprise alpha. Audio is deleted after notes are ready. OpenAI instructs users to obtain participants’ consent before starting. Source.
Shareable profiles
Profiles offer a place to showcase creations and activity. People can choose Sites to display, while personal and Business profiles have separate sharing controls. The Help Center says Enterprise profile sharing is coming soon, despite its inclusion in the recap’s broader plan list. Source.
Subscriptions and the enterprise ecosystem
Sign in with ChatGPT
People can sign into participating products with their ChatGPT account. Plus and Pro users can also allow eligible requests to use their existing plan allowance and set per-app weekly limits. Partner products may charge separately for their own services; this does not make every participating app free. Source.
Pro 500 and the return of Pro 200
The new Pro 500 tier costs $500 a month and includes Ultrafast. OpenAI describes its allowance as 25 times Plus. Pro 200 is open to new subscriptions again at $200 a month, with a lower allowance than the earlier version. Eligible existing subscriptions keep their old allowance through October 29, 2026. Source.
OpenAI Marketplace
Eligible enterprise customers can spend part of an existing OpenAI commitment on approved partner software. The marketplace includes companies such as Figma, Adobe, Notion, Harvey and Baseten. It extends enterprise purchasing beyond OpenAI’s own products. Source.
Other keynote updates
OpenAI also reported a 45% reduction in API time to first token during the keynote. That is a company-reported service improvement, rather than a new model. The opening recapped earlier Codex additions including Linux support, multiple folders in a project and phone access. Watch the keynote.
The direction behind the announcements
Our reading of DevDay is that OpenAI wants the model, the working environment, the connected tools and the shared output to sit much closer together. Dots carry ongoing context. Space gives the work a home. Plugins and APIs let other software participate. The immediate test is whether those pieces make a real task easier to start, continue and finish.