OpenAI used DevDay 2026 to introduce a new model, persistent personal agents, cloud-based development environments, shared team automations, meeting capture, and new ways to connect websites with ChatGPT.
The announcements cover different products, but they move in a consistent direction: ChatGPT is expanding from a conversational assistant into a system that can perform ongoing work across applications, devices, teams, and web experiences.
Many of the features are rolling out gradually. Availability varies by plan, region, device, and workspace policy.
GPT-6.1 Sol brings near-Astra performance to more work
OpenAI introduced GPT-6.1 Sol, the newest model in the GPT-6 family.
The company positions it between GPT-6 Astra and GPT-6 Luna: designed for complex coding, computer use, and professional work, but at a lower cost than Astra. The model supports a context window of more than one million tokens and can use tools including web search, computer use, code execution, image generation, MCP, and skills.
GPT-6.1 Sol is rolling out through ChatGPT Work and Codex on supported paid plans. It is also available through the API, but it is not currently available as a model in standard Chat. OpenAI’s model documentation → provides the technical specifications and pricing.
The practical implication is that more demanding agentic work may no longer require the most expensive frontier model. Teams can select among Astra, Sol, and Luna based on the reasoning, speed, and cost required for a particular workflow.
Dots are persistent personal agents
One of the largest product announcements was dots: personal agents that can remain active across conversations and tasks.
A dot is powered by GPT-6 Astra and has its own cloud computer. It can use connected applications, retain context for ongoing work, and create tasks in ChatGPT Work or Codex. Depending on the permissions enabled, it can also interact through Slack or Microsoft Teams and use a connected local computer.
Unlike a conventional ChatGPT conversation, a dot is designed to persist. It can remember relevant context, monitor connected information, and continue working through different channels.
Dots are beginning to roll out to Pro users outside the European Economic Area, Switzerland, and the United Kingdom. OpenAI is also offering a beta for eligible Enterprise workspaces, where access is disabled by default and controlled through administrative permissions. OpenAI’s dots documentation → explains the initial availability and capabilities.
This introduces a more continuous form of AI assistance, but it also makes permissions, memory, identity, and action controls central parts of the experience.
Codex Cloud moves development work off the local computer
OpenAI also announced new Codex Cloud capabilities for Enterprise.
Codex Cloud runs coding tasks on OpenAI-managed computers. Organizations can prepare reusable environments containing the repositories, tools, dependencies, and access settings required for a project.
Once an environment is available, authorized users can start or continue tasks from desktop, web, or mobile. The task runs in its own isolated workspace and can continue while the person’s computer is offline or asleep.
The distinction is important. Codex is no longer limited to assisting someone while they work locally. A developer can delegate work to a cloud environment, leave it running, and return later to review the result.
Access remains subject to workspace settings and repository permissions. OpenAI notes that some of the DevDay features are still rolling out and may not yet appear in every account. The Codex Cloud guide → provides the current availability and setup details.
Team tasks turn automations into shared work
Scheduled tasks previously focused primarily on work maintained by an individual. The new team-task capabilities allow multiple authorized people to manage recurring work together.
A team task can run on a schedule or supported trigger, use approved tools and workspace connections, and make its results available to the team. Authorized teammates can review the output, revise the instructions, and maintain the task as requirements change.
OpenAI gives examples including daily customer-change reports, weekly project updates, and meeting briefings assembled from selected documents and recent activity.
The key change is ownership. A recurring AI workflow can become a shared team resource instead of remaining attached to one person’s account or prompt history. OpenAI’s team-task documentation → describes the initial workflow.
Teams can share approved tool connections
OpenAI is also adding centrally managed connections for team workflows.
Enterprise administrators can configure approved application connections for the workspace. Team owners can then enable eligible connections for their teams, reducing the need for every employee to connect the same provider account separately.
Individual permissions and provider restrictions still apply. Installing a plugin or enabling a team connection does not give ChatGPT unrestricted access to the underlying service.
This is an operational improvement, but it is also a governance feature. Organizations gain a more consistent way to define which tools AI workflows can use and how those tools are made available across teams.
The controls are outlined in OpenAI’s Enterprise and Edu release notes →.
The Meetings plugin connects conversation to follow-up work
The new Meetings plugin can capture audio from online or in-person conversations, create notes, identify action items, and save personalized summaries in ChatGPT Space.
Because the result becomes part of ChatGPT, the notes can then be used in later research, planning, writing, or task workflows.
The plugin is currently in beta in the ChatGPT desktop application for macOS. It is available on Pro and Business plans, while Windows and mobile support are planned. It is not yet generally available for Enterprise.
OpenAI says the meeting audio is deleted after the notes are prepared and cannot be replayed. The Meetings plugin guide → includes the current availability, privacy details, and limitations.
ChatGPT Sites can now host tools, not just interfaces
DevDay also expanded the role of ChatGPT Sites.
A Site can now host an MCP server and expose its tools through a ChatGPT plugin. This allows a website to serve two related functions: provide a visual interface for people and provide structured capabilities that ChatGPT or Codex can use.
For example, a team handbook could include tools for searching company policies. A project dashboard could allow an authorized agent to read milestones or update status. The Site remains the visible destination, while the plugin makes selected capabilities available through conversation.
Hosting a plugin through Sites is available across plans. Business and Enterprise workspaces may require administrator approval before a Site can be published or its plugin shared. OpenAI’s Sites plugin documentation → explains how the Site, MCP tools, and plugin work together.
For web and digital teams, this is an important development. Websites can increasingly be designed for both direct human use and agent-mediated interaction.
OpenAI Marketplace adds an enterprise purchasing channel
OpenAI also announced the OpenAI Marketplace in beta for eligible enterprise customers.
The program allows participating organizations to apply part of their OpenAI commitment toward qualifying software from approved partners. Eligibility is determined by OpenAI for each customer, product, and purchase.
The announcement does not create an unrestricted public marketplace. It is currently an enterprise program managed directly with OpenAI and participating vendors.
What the announcements mean together
The DevDay releases span models, agents, cloud environments, meetings, team operations, websites, plugins, and enterprise administration.
Taken together, they indicate several changes:
- AI work is becoming more persistent. Tasks and agents can continue beyond an individual conversation or device.
- AI workflows are becoming shared organizational resources. Teams can manage tasks, environments, connections, and plugins together.
- Websites are gaining a second interface. People can use the visible experience while agents interact with structured tools behind it.
- Governance is becoming part of the product. Access, approvals, identity, memory, data connections, and action boundaries increasingly determine how the system operates.
- The model is only one component. The surrounding environment, tools, permissions, and workflow design determine what an agent can actually accomplish.
OpenAI’s latest releases do not replace the applications, websites, and systems organizations already use.
They create a new layer that can work across them.
Did you like this?