SAN FRANCISCO, United States, September 29, 2026: OpenAI used its DevDay 2026 conference in San Francisco to unveil one of its broadest product expansions yet, pushing ChatGPT further into workplace collaboration, autonomous agents, software development, identity, cybersecurity and enterprise infrastructure.
The announcements stretched well beyond a new model release.
OpenAI introduced GPT-6.1 Sol, expanded access to GPT-6 Astra, launched Dots as persistent AI agents, added Teams, Pages and broader sharing inside ChatGPT, brought @ChatGPT deeper into workplace messaging, expanded Codex into cloud development and security, opened new developer pathways around plugins and the Model Context Protocol, and introduced new privacy controls for organisations using the API.
Okay News reports that many of the products announced at DevDay are rolling out gradually, while others are available only in beta, research preview or limited preview.
The scale of the release points to a larger shift in OpenAI’s product strategy. ChatGPT is increasingly being positioned less as a single conversational interface and more as an operating layer connecting people, agents, files, software repositories, workplace applications and external services.
At the centre of that shift is GPT-6.1 Sol.
OpenAI introduced GPT-6.1 Sol as the latest model in the GPT-6 family, designed to deliver capabilities close to its most powerful model, GPT-6 Astra, while improving the balance between speed and cost.
The model is being deployed in ChatGPT Work and Codex, where it is intended for coding, computer use and professional tasks that require sustained reasoning across multiple steps.
GPT-6 Astra remains OpenAI’s highest-capability model for the most demanding workloads. It was introduced earlier in September with stronger performance in software engineering, browser use, research, cybersecurity and complex knowledge work.
DevDay extended that story with Ultrafast, a lower-latency mode for GPT-6 Astra aimed at users who need frontier capability with faster response times.
At launch, Ultrafast is available on the new ChatGPT Pro $500 tier and in eligible Enterprise and Edu workspaces.
OpenAI has kept the distinction between capability and speed clear. Ultrafast does not represent a different model from Astra. It is a faster execution mode, with usage priced or deducted at a higher rate because of the additional compute required to reduce latency.
The new Pro $500 tier therefore sits above the company’s existing Pro offerings and is aimed at users with unusually heavy workloads in ChatGPT Work and Codex.
Another major announcement was Dots.
Dots are persistent AI agents that can keep working between conversations rather than waiting for a user to open a chat and issue another prompt.
Each dot is powered by GPT-6 Astra and can have its own cloud computer, connected applications, goals and operating rules.
A user can assign ongoing responsibility to a dot, define what it is allowed to do independently and allow it to continue making progress between interactions.
The agent can then return for decisions that require human judgement.
This moves ChatGPT closer to the idea of an always-on digital colleague rather than a session-based assistant.
Dots are initially rolling out to Pro users in supported markets, Business Premium customers and selected Enterprise organisations through a beta controlled by workspace administrators.
Enterprise administrators can manage whether dots are available, whether they can connect to workplace messaging systems and whether they can use local computer resources.
OpenAI also extended the agent model into Slack and Microsoft Teams.
Organisations can now deploy @ChatGPT into approved messaging environments, allowing employees to ask questions, summarise conversations, draft material and work with information exposed through connected company tools.
The integration can use approved organisational context without requiring each user to separately configure every shared provider connection.
Dots can also be connected to supported Slack and Microsoft Teams environments, although access depends on organisational permissions and rollout status.
The distinction matters.
@ChatGPT functions as the organisation’s shared assistant surface, while a dot belongs to an individual user and can take on continuing responsibility on that person’s behalf.
DevDay also introduced a Meetings plugin, currently in beta.
The tool can listen to online or in-person meetings through a user’s computer, generate personalised notes and action items, and save the results into ChatGPT Space.
Unlike meeting assistants that join calls as visible bots, OpenAI’s Meetings plugin can work through microphone and system audio from the user’s Mac.
Users must still obtain consent from other participants before recording or processing meeting audio.
Meeting notes can incorporate context from connected applications and can become the starting point for follow-up work, such as drafting an email, updating a project plan or assigning another task to ChatGPT.
At launch, Meetings is available in beta on macOS for Pro and Business users, with broader platform support planned.
The meeting workflow connects directly with another major set of changes announced around ChatGPT Space, Pages and Teams.
ChatGPT Space is becoming the home for files and Pages.
Pages are editable documents that can be organised inside other Pages and collaboratively worked on with ChatGPT.
Users can share individual Pages or entire Spaces and give collaborators viewing or editing access.
Each person working on a shared Page can still use their own ChatGPT agent against the same material.
OpenAI also introduced Teams inside ChatGPT, separate from Microsoft Teams.
A ChatGPT Team is a group of coworkers inside the same Business or Enterprise workspace who repeatedly work on the same projects.
Instead of sharing the same Space or Page individually with multiple colleagues, users can create a Team and share resources with the group.
Teams can also jointly manage recurring work.
A team task can run on a schedule or supported trigger, with authorised teammates able to review results and update the instructions over time.
This effectively brings team automations into the collaboration layer.
A marketing team could maintain a recurring campaign update. A customer-success team could run a weekly account digest. A product group could monitor feedback and produce a regular summary.
The same resource can be shared with more than one team, while a team can work across multiple Spaces.
OpenAI also introduced shareable profiles.
Users can customise a ChatGPT profile with a name, username, photo, biography, activity and selected Sites they want to showcase.
Personal profiles are private by default but can be shared publicly where supported.
Public profile links use a ChatGPT username and can showcase Sites built by the user.
The feature gives ChatGPT a lightweight identity and portfolio layer, particularly for developers, creators and professionals publishing work through OpenAI’s ecosystem.
OpenAI is making that identity portable outside ChatGPT through another DevDay announcement: Sign in with ChatGPT.
The feature allows participating websites and applications to offer a “Continue with ChatGPT” option alongside other identity providers.
Users can create or link an account using their ChatGPT identity.
For ordinary identity sign-in, the external service receives basic profile information such as the user’s name, email address and profile picture if available.
Chat history, Memory, files and billing information are not automatically shared.
A separate permission can allow eligible users to use their ChatGPT plan allowance inside partner applications.
That could remove one of the biggest adoption barriers for smaller AI applications.
Instead of forcing users to obtain API keys or immediately pay for another AI subscription, participating applications can allow eligible Plus and Pro users to sign in with ChatGPT and use supported AI workloads against their existing plan allowance or credits.
The initial Sign in with ChatGPT rollout includes partners such as Airtable, GitLab, HubSpot, Notion, Supabase and Vercel, while commercial partner access remains limited and subject to approval.
For open-source applications, OpenAI has also published a path that lets eligible users bring their own ChatGPT plan without manually configuring an API key.
Software development received some of the largest changes announced at DevDay.
OpenAI launched Codex Cloud, giving developers reusable cloud environments containing their repositories, tools and dependencies.
A developer can begin a task on desktop, web or mobile and allow it to continue in an isolated cloud workspace even when the developer’s own machine is asleep.
This makes Codex less dependent on a continuously running local development environment.
Enterprise users can also share approved cloud environments, allowing teams to standardise dependencies and tooling.
The Codex CLI has also been refreshed, with expanded agent workflows and voice interaction.
Developers can work with Codex through natural language in the terminal, invoke or coordinate agents and move between local and cloud development workflows.
The broader Codex system now includes formal Code Review and automatic review workflows.
OpenAI’s credit documentation distinguishes conventional code review from auto review, showing that review jobs are now part of the supported Work and Codex usage model.
The company is also broadening repository support, including GitLab, reducing the earlier dependence on GitHub-only workflows for some development tasks.
Cybersecurity received its own expansion through Codex Security.
Codex Security is designed to work more like an application security researcher than a conventional static scanner.
It builds a threat model of a connected codebase, examines potential attack paths, attempts to validate vulnerabilities in an isolated environment and proposes patches for human review.
The system can scan full repositories or commits and tries to reproduce suspected vulnerabilities before surfacing them, which OpenAI says is intended to reduce false positives.
Codex Security does not automatically modify production code.
Validated findings can result in proposed patches that teams can review and convert into pull requests through their existing workflow.
The product remains in research preview for eligible ChatGPT Business, Enterprise, Edu and Pro customers.
OpenAI also brought computer use into the Agents API, allowing developers to create agents that operate an OpenAI-hosted browser.
An agent can navigate websites, interact with graphical interfaces and complete browser tasks while the developer’s application follows session events.
The API uses explicit controls around website access and approvals.
Developers can inspect the browser session, approve access to new origins and handle sign-ins where required.
The capability extends the Agents API beyond tool calls and text generation into direct interaction with interfaces that were not built specifically for AI agents.
That agent infrastructure is also moving into Amazon Web Services.
Amazon Bedrock Managed Agents powered by OpenAI entered preview following joint work between AWS and OpenAI.
The service uses a customised OpenAI agent runtime integrated into AWS infrastructure.
It manages state, tool use, code execution and multi-step workflows while working with AWS identity and governance controls.
Agents can use their own IAM roles, connect to Model Context Protocol servers and support human approval before consequential actions.
The preview initially runs in selected United States AWS regions.
OpenAI also announced a Decisions API in limited preview, expanding its developer platform for systems that need AI to reason over choices and return structured decisions.
That sits alongside a wider push to make OpenAI infrastructure more useful for long-running agents, automated workflows and enterprise systems rather than individual prompt-response interactions.
Plugins received another round of expansion.
OpenAI is extending plugin interfaces beyond basic chat responses through richer sidebars, panels and file viewers, allowing third-party tools to present more useful interfaces directly within ChatGPT.
Developers can use the Apps SDK, which extends the Model Context Protocol, to define both application logic and interactive user interfaces.
OpenAI is also accepting submissions to the ChatGPT plugin directory, with changes intended to improve how users discover third-party tools.
The directory is becoming a central distribution channel for applications that can work across ChatGPT and Codex.
OpenAI has also expanded the underlying Model Context Protocol infrastructure through MCP Events.
MCP Events allow ChatGPT to subscribe to events coming from an external system and act when something changes.
A user could ask ChatGPT to monitor a product-feedback channel for bug reports, watch a document for comments or respond when a supported business system emits another event.
The system uses webhook delivery rather than repeatedly polling a service for updates.
This is the infrastructure that makes event-driven automations possible.
A recurring task no longer has to ask an external system every hour whether something changed. The external MCP server can inform ChatGPT when the change actually occurs.
OpenAI is also broadening developer support around MCP extensions and SDKs, including Python and Node tooling, to make plugin development easier and more consistent.
The company’s direction is to make MCP a shared connective layer between AI applications, tools and external services rather than requiring every integration to use a proprietary format.
Developers can also use Sites to host interactive applications and experiences built with ChatGPT.
A project created through ChatGPT can be deployed and shared through a live Site, allowing users to move from a conversation into an interactive application without separately building a conventional deployment pipeline for every prototype.
The combination of Sites, plugins, Apps SDK interfaces and MCP support means developers can increasingly build an experience, connect tools, deploy it and distribute it from within the same ecosystem.
DevDay also included changes aimed at organisations with strict data-retention requirements.
OpenAI introduced Zero Data Retention with Private Safety Processing, or ZDR with PSP.
The system is intended to allow automated safety monitoring without requiring OpenAI to retain readable copies of customer prompts and model responses.
Eligible content selected for safety review is encrypted and stored in customer-controlled cloud storage.
Automated safety processing occurs inside a hardware-attested environment designed to prevent human access to the protected content.
Only bounded safety signals and approved operational metadata can leave the protected environment in plaintext.
Customers can use storage in AWS, Microsoft Azure or Google Cloud, subject to configuration and eligibility requirements.
The design is particularly relevant for regulated organisations that want to use more capable models while maintaining strict control over where sensitive data is stored.
Taken together, DevDay 2026 showed OpenAI pushing in several directions at once.
The company is making its frontier models faster and more accessible through GPT-6.1 Sol and Astra Ultrafast.
It is making agents more persistent through Dots and more collaborative through Teams, Spaces and shared automations.
It is pulling ChatGPT deeper into Slack, Microsoft Teams and meetings.
It is turning Codex into a cloud development and security environment.
It is allowing users to carry their ChatGPT identities and plan usage into external applications.
And it is giving developers more ways to connect outside systems through plugins, MCP and the Agents API.
The common thread is that OpenAI is moving beyond a product centred on individual conversations.
The new system assumes AI will increasingly sit inside ongoing work: monitoring projects, operating software, collaborating with teams, reviewing code, attending meetings, running in cloud environments and communicating across services.
Some of what OpenAI announced is already broadly available. Other features remain limited to particular plans, regions, organisations or previews.
That distinction will matter as developers and companies decide what can be deployed immediately and what still needs to be treated as experimental.
But DevDay 2026 made OpenAI’s direction considerably clearer.
ChatGPT began as somewhere users went to ask an AI a question.
OpenAI is now building toward a world where the AI stays with the work after the question is over.
