At DevDay 2026 on September 29, OpenAI announced 25 changes spanning persistent agents, a new model, developer tools, shared work and subscriptions. Dots are rolling out and GPT-6.1 Sol is available in selected surfaces; collaborative slides and private inference are still ahead.
Dots, Models and Speed
Dots are persistent agents powered by GPT-6 Astra. Each has a cloud computer and can work across connected apps while the user is away. Their idle proactive research uses read-only tools; actions affecting accounts or sharing information face auto-review, custom rules and approval controls. Finding work and carrying it out are different permissions.
Dots are rolling out to eligible Pro users outside the EEA, UK and Switzerland, including Pro 100, and to Business Premium users in supported markets. Enterprise, Edu and Healthcare access is an admin-enabled beta, off by default. For the first month, Dots usage does not count toward eligible Pro, Business and Enterprise plan allowances. Usage limits after the first month remain unspecified. Specialist dots with separate organizational identities are in enterprise pilots.
GPT-6.1 Sol follows last week's GPT-6 Sol and Luna launch. OpenAI's evaluations put it near Astra on coding, computer use and professional work. It is available now as gpt-6.1-sol in the API and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu, but not yet in ordinary Chat. Standard API rates are $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens, one-fifth of Astra's standard input and output rates. The cached-input rate matters for agents reusing context; total cost still depends on the full workload.
Ultrafast is the premium speed tier. Astra Ultrafast is available now in the API with US data residency or global processing, and in Work and Codex for Pro 500 and eligible Enterprise and Edu plans. Enterprise owners must enable it; legacy rate-limit Enterprise plans and non-US inference residency are unsupported. Codex offers up to eight times faster token generation, which does not measure complete task time. GPT-6.1 Sol Ultrafast is coming soon.
Private Intelligence has a split timeline too. Zero Data Retention with Private Safety Processing provides automated safety review for eligible API organizations without OpenAI retaining prompts or responses. Private Inference, combining confidential computing with verifiable controls, is a fall preview.
Codex and the Developer Stack
Codex in the cloud extends work beyond a single computer, with reusable development environments and shared approved settings. A refreshed Codex CLI adds voice steering and an /agents view for parallel tasks. Desktop Code Review brings diffs and issue discussion together, with optional automatic cloud reviews. GitHub review is generally available; GitLab merge requests are in preview. A review suggestion still needs human judgment before becoming a code change.
Codex Security Cloud scans GitHub repositories on demand or on a schedule, investigates findings and prepares fixes in the cloud. It supports Pro, Business, Enterprise and Edu on desktop and web. The narrower Decisions API uses Luna to choose among developer-defined answers for classification or routing; it is in limited preview, with broader release planned in the coming days.
The existing Agents API adds computer use alongside its multi-agent, tool-search and context-compaction capabilities. It is available through the API, with Work and Codex access for Pro 500 and Enterprise. Amazon Bedrock Managed Agents offers an AWS-native route for OpenAI-powered agents in limited preview. Computer use makes permission boundaries more consequential.
Plugins Become More Like Apps
Plugin extensions give developers sidebar homes, interactive panels and custom file viewers inside ChatGPT. Their web rollout for Free and Go is coming soon, while composer mentions work only in the desktop app. Plugin Creator, a clearer submission flow and improved discovery ease building and finding plugins. Users choose them and approve their access.
Sites can now host supported plugins for teammates, each using their own connected data and permissions, on Business, Enterprise, Healthcare and Edu. MCP events let a connected app trigger an automation when something changes. Support follows a proposed specification; it is not universal across MCP servers.
Shared Work, Meetings and Profiles
ChatGPT Space gives Pro, Business and Enterprise users a shared home for people, ChatGPT and Dots. Its Pages are collaborative documents where a team can write, research and build with agents. Creating and editing Space content is on desktop and web; mobile can find, read and share pages, with mobile editing to come. Collaborative slides, including simultaneous teammate and agent editing, are due in the coming weeks rather than available today.
Business and Enterprise gain teams and shared tasks for recurring or event-triggered work, plus @ChatGPT in Slack and Microsoft Teams using connected tools and user permissions. The Meetings plugin is a macOS desktop beta for Pro and Business that saves notes and action items into Space; Enterprise support comes later. Shareable profiles collect Sites and plugins, while shared skills remain within a workspace. Enterprise profiles are coming soon.
Plans and Partner Software
Sign in with ChatGPT separates account identity from plan usage. Identity works globally at participating apps; eligible Plus and Pro users can also allow supported apps to draw on their ChatGPT Work and Codex allowance. That usage shares the plan's limits, can have a weekly cap per app, and does not erase any separate partner charge. The launch line-up includes 16 plan-usage partners; a ChatGPT login elsewhere does not imply those benefits automatically.
The new Pro 500 costs $500 a month and includes Astra Ultrafast. New Pro 200 subscriptions have reopened after the earlier pause at the same $200 monthly price. Buying credits on Pro 100 or Pro 200 does not unlock Ultrafast.
Shared Work and Codex usage allowances now use Plus 1X, Pro 100 5X and Pro 200 10X. The earlier Pro 200 Codex weekly allowance was 20X relative to the old Plus baseline. Plus's baseline is rising too, so the ratios do not measure completed tasks. At API rates, the new Pro 200 allowance has half the former dollar value. Eligible existing subscribers retain the previous allowance through October 29, 2026, then move to the lower one. OpenAI has also promised extra credits, with no stated amount or duration, and no return of the five-hour cap.
The OpenAI Marketplace lets eligible enterprises apply part of an existing OpenAI commitment to approved partner software. Customers can express interest; access is not automatic.
What do I think of this?
Persistent agents are the headline change for me. Dots can continue work across already-connected apps, Codex can run away from a laptop, and Space keeps team material in one place.
The Pro 200 allowance is the weak point: new heavy users pay the same $200 for less included Work and Codex usage, and eligible existing users face the reduction after October 29. Cheaper models and unmetered features do not prove every workload will recover that lost allowance. One month of exempt Dots usage and extra credits without a stated amount or duration do not settle longer-term value.