OpenAI used its September 29, 2026, DevDay conference to introduce more than 20 announcements across ChatGPT, Codex, its models, and developer tools. CEO Sam Altman opened the event, outlining the company's latest developments and its direction for AI-powered workflows.
One of the key announcements was OpenAI Dots, an always-on AI agent designed to handle ongoing tasks across applications. Alongside Dots, updates to the Agents API and Codex give developers more ways to build, coordinate, and manage AI-powered workflows. Together, these announcements reflect OpenAI's focus on AI systems that can do more than respond to individual prompts.
OpenAI Dots Bring Always-On AI Agents to ChatGPT
Dots were the headline announcement at DevDay. Unlike a conventional chatbot that waits for a user to send a message, Dots are designed to remain active and take responsibility for ongoing tasks. OpenAI describes them as personal agents that learn what matters to users and work in the background to help manage their responsibilities.
Each Dot has its own cloud computer and can connect to more than 4,000 apps. Users can interact with their agents through ChatGPT, Slack, and Microsoft Teams, with text messaging planned for a later release. OpenAI also says Dots can learn from feedback over time, allowing users to refine how they work.
The idea is to move beyond asking an AI assistant to complete a single task. A user might assign an ongoing responsibility, provide relevant context, and allow the agent to help manage the work as circumstances change. OpenAI demonstrated potential applications for developers, scientists, executives, sales teams, and content creators.
For example, a content creator could use an agent to help organize research and track an editorial workflow. A developer might delegate recurring tasks related to a project. These are examples of the product's intended uses, not guarantees that every task can be completed without supervision.
OpenAI also plans to support teams of Dots working together in the future. At launch, users begin with a primary Dot that they can name and customize. The company has not yet made the full multi-agent experience generally available.
Dots are rolling out to ChatGPT Pro and Business Premium users in eligible markets. Enterprise, Edu, and Healthcare users can access a beta when their workspace administrator enables it. This staged rollout means availability depends on both the subscription and the workspace settings.
The launch also raises practical questions about how users should delegate work to an always-on agent. Connecting an agent to apps and giving it ongoing responsibilities can make it more useful, but it also makes permissions, oversight, and clear boundaries important. OpenAI's rollout and product controls will be worth watching as more users begin working with Dots.
GPT-6.1 Sol Expands OpenAI's Model Lineup
OpenAI also introduced GPT-6.1 Sol, an upgrade to GPT-6 Sol. The company says the model delivers strong performance in agentic coding, computer use, and professional work. It is positioned as a model that can handle demanding workflows while costing less than GPT-6 Astra.
According to OpenAI, GPT-6.1 Sol offers near-Astra intelligence at one-fifth of Astra's standard input and output token prices. The lower cost could give developers more flexibility when building applications that make repeated model calls or require agents to work through multi-step tasks.
That pricing distinction matters because agent-based applications can involve more than one request. An agent may need to interpret instructions, use a tool, inspect the result, and decide what to do next. Each step can consume tokens. A lower-cost model may therefore affect the economics of running an application, although the actual savings will depend on the workload and how efficiently it is designed.
GPT-6.1 Sol is available to API, ChatGPT Plus, Pro, Business, Enterprise, and Edu users, according to OpenAI. Developers evaluating the model should test it against their own tasks rather than assuming that a general performance claim will translate directly to every application.
The model announcement came shortly after OpenAI decided not to release GPT-6.1 Astra because it did not meet the company's safety standards. The distinction is important: GPT-6.1 Sol is the model OpenAI introduced at DevDay, while the planned Astra release was held back. The two should not be treated as interchangeable products.
Ultrafast Adds a Premium Speed Tier
For users who prioritize response speed, OpenAI introduced Ultrafast, a premium tier for workloads where latency matters. The company says it can generate tokens up to eight times faster in Codex and up to six times faster in the API.
Faster generation may be useful for developers working interactively with code or applications where users are waiting for an AI response. It may also help workflows that involve repeated model interactions. However, speed is only one part of the decision. Developers still need to consider cost, output quality, reliability, and whether a particular task benefits from a faster response.
GPT-6 Astra Ultrafast is available in the API and for eligible ChatGPT Work and Codex users on Pro 500 and Enterprise plans. OpenAI says GPT-6.1 Sol Ultrafast is coming soon. The distinction between the available and upcoming versions matters for teams planning to adopt the feature.
Codex and the Agents API Give Developers More Ways to Build
While Dots are aimed at people who want an agent to take on ongoing responsibilities, OpenAI's developer announcements focus on building applications that can use agents to complete work. Updates to Codex and the Agents API are part of that broader effort.
The Agents API gives developers tools for creating AI agents that can use software and work through more complex tasks. OpenAI introduced computer-use capabilities that allow agents to interact with software through its interface. The company also announced features such as multi-agent orchestration, tool search, and context compaction.
These capabilities address different parts of agent development. Computer use lets an agent interact with applications, while orchestration helps coordinate multiple agents. Tool search can help an agent find relevant tools, and context compaction is intended to manage the information an agent carries through a longer task.
For developers, the significance is not simply that an agent can generate an answer. It is that an application can be designed to perform a sequence of actions, use tools, and handle more of a workflow. That also creates a need for careful testing, especially when an agent can interact with external systems or modify data.
Codex received updates of its own, including cloud-based development capabilities and security-related tools. OpenAI also introduced Codex Security Cloud, which can scan code repositories and help investigate vulnerabilities. These features extend Codex beyond code generation into parts of the software development and review process.
Developers will still need to review generated code and validate security findings. AI-assisted tools can help identify issues, but their output should be treated as part of a development workflow rather than a replacement for engineering review.
The Decisions API Targets Structured AI Decisions
Another developer announcement was the Decisions API, which is designed for tasks where an application needs a response from a defined set of possible choices. Instead of generating a long, open-ended answer, the system can help classify an input, route a request, or select an action from a predefined set.
This type of capability can be useful in applications that need consistent outputs. For example, a support system might need to route a request to a particular team, while another application might classify incoming information before passing it to a different service.
The Decisions API is in limited preview, so developers should check its availability before planning a production deployment. Its value will depend on how well it performs on the application's specific decision-making tasks and how the developer handles uncertain or incorrect outputs.
ChatGPT Space and Pages Bring People and Agents Together
OpenAI's announcements were not limited to individual agents and developer infrastructure. The company also introduced collaboration features intended to bring people and AI into shared workspaces.
ChatGPT Space allows teammates and Dots to work together in a shared environment. This gives teams a place to collaborate with AI agents rather than treating every interaction as a separate conversation. OpenAI also introduced Pages, a document format designed for people and agents to work together on writing, images, charts, and visualizations.
These features point toward a more collaborative use of AI. Instead of copying an answer from a chatbot into a separate document, users can work with AI within a shared space and develop the result alongside colleagues.
For teams, the practical value will depend on how well these features fit existing workflows. Organizations will need to consider access permissions, document ownership, review processes, and how AI-generated content is checked before it is shared or published.
Private Intelligence and Enterprise AI
OpenAI also previewed Private Intelligence, an offering designed to give organizations stronger controls over their data while using its models. The announcement included Zero Data Retention with Private Safety Processing and a planned Private Inference preview.
These capabilities are relevant to organizations that have strict requirements for handling sensitive information. OpenAI said it has worked with large customers, including Cisco, Databricks, and Snowflake, on privacy and safety systems.
However, the announcement should not be read as a blanket guarantee that every enterprise use case is automatically compliant with every organization's requirements. Businesses still need to review the available controls, contractual terms, data flows, and applicable policies before adopting an AI service for sensitive workloads.
OpenAI also announced additional enterprise and platform features, including an OpenAI Marketplace and ways for developers to build more native experiences inside ChatGPT. These updates expand the company's efforts to make ChatGPT a place where users can access applications and work with AI-powered tools.
What OpenAI's DevDay Announcements Mean for AI Development
The announcements at OpenAI's DevDay, opened by CEO Sam Altman, share a common direction: making AI more useful across longer and more complex workflows. Dots are designed to handle ongoing responsibilities for users, while the Agents API gives developers tools to build agents into their own applications. Codex updates address software development, and ChatGPT Space and Pages bring AI into collaborative work.
The model and infrastructure announcements support that direction. GPT-6.1 Sol offers another option for developers balancing capability and cost, while Ultrafast targets workloads where speed is especially important. The Decisions API focuses on structured choices that can be integrated into applications.
But more capable agents also introduce new questions. An agent that can access applications and act over time needs clear permissions, reliable monitoring, and ways for users to understand what it has done. The decision to hold back GPT-6.1 Astra over safety concerns adds context to OpenAI's expansion into more autonomous systems.
For developers and businesses, the announcements are best understood as a set of tools to evaluate rather than a reason to redesign existing workflows immediately. The right starting point will depend on the task: some teams may benefit from a lower-cost model, while others may be interested in computer use, coding support, or shared AI workspaces.
Conclusion
OpenAI DevDay 2026 introduced a broad set of changes across models, agents, developer tools, and collaboration features. Dots received the most attention, but the wider announcement matters because it connects personal agents with the infrastructure developers need to build and deploy AI-powered workflows.
The next stage will depend on how these products perform outside demonstrations, how quickly availability expands, and whether users can rely on agents to operate within clear boundaries. For now, DevDay offers a view of OpenAI's direction: AI systems that do not just answer questions, but can help people and teams carry work forward.
