OpenAI DevDay 2026: ChatGPT becomes a collaborative workspace and agent platform

On 29 September 2026, OpenAI held DevDay at Fort Mason in San Francisco. The headline was GPT-6.1 Sol, a meaningful upgrade to the existing Sol model, but the more consequential shift was structural: ChatGPT stopped being a chat interface and started being a shared workspace. Here is what shipped and what it means for your work.

ChatGPT Space and Pages: a shared workspace for teams and agents

ChatGPT Space is a new shared environment where your team, ChatGPT, and OpenAI’s Dots agents all work from the same context. Think of it less like a chat thread and more like a lightweight project hub. It sits alongside @ChatGPT integrations in Slack and Microsoft Teams, so you can pull it into the tools you already use.

The centrepiece of Space is Pages, a new document type designed for humans and agents to build together. Pages support writing, image generation, charts, and data visualisation. If you have used Notion with an AI layer baked in rather than bolted on, you will have a reasonable mental model. Collaborative slides are due in the weeks following the event.

Space is available on Pro, Business, and Enterprise plans on desktop and web. Mobile supports reading and sharing Pages from launch, with creation and editing to follow.

What does this mean for you? If your team is currently copying AI outputs into documents manually, Space removes that step. Agents can draft, you can edit, and everything stays in one place rather than bouncing between tabs.

Plugin Extensions: full applications inside ChatGPT

OpenAI is opening the platform it uses to build its own ChatGPT features to external developers. Plugin Extensions let developers add a sidebar destination, interactive panels, file viewers, and forms directly inside ChatGPT. These are not simple chat plugins. The keynote demonstrated Figma designs and comments appearing alongside a conversation, Photoshop functionality embedded in the interface, and OpenAI’s own Meetings app running as a native panel.

The practical result: applications that previously required switching windows can now sit inside ChatGPT and share the conversation’s context. OpenAI distributes these through its own platform, so users find them in one place.

For developers, this is a meaningful distribution opportunity. Your tool gains access to 1.2 billion weekly ChatGPT users without requiring them to sign up for a separate product.

Codex cloud tasks: work that continues when your laptop is closed

Codex can now run tasks in the cloud rather than on your local machine. Close your laptop, and the task keeps going. Check progress from your phone. Start a new task from anywhere.

Reusable environments are the other practical addition here. A team can define one approved setup, including permissions and configuration, and Codex will start new tasks in isolated workspaces derived from that shared baseline. No more “works on my machine” problems and no dependency on a single developer’s laptop staying online.

This is available on Plus, Pro, Business, Enterprise, Edu, and Healthcare plans.

Decisions API: faster classification without full LLM overhead

The Decisions API is in limited preview from 29 September 2026, with a broader release planned shortly after. It is built on a version of GPT-6 Luna and is designed for a specific problem: using a full large language model to classify content or route a request is expensive and slower than it needs to be.

The Decisions API takes context in text or image form, applies it to a pre-defined set of questions with finite answers, and returns a classification. Use cases include content moderation, request routing, and selecting an agent’s next action. It is not a replacement for general-purpose model calls; it is a faster, cheaper tool for decisions where the answer space is bounded.

Sign in with ChatGPT lets your ChatGPT plan’s usage allowance cover eligible activity inside 16 partner tools, including Cognition’s Devin, Notion, and Vercel. You control how much each tool can draw from your allowance.

For individual developers building AI-powered products, this matters. Rather than requiring users to bring their own OpenAI API key or absorbing API costs yourself, eligible usage can count against the user’s existing ChatGPT plan. That removes a meaningful barrier to building and shipping.

MCP Events: event-driven automation without polling

OpenAI is adding support for the proposed MCP Events specification. Until now, MCP usage in ChatGPT was primarily about AI calling tools on demand. MCP Events inverts that: something happens in a connected external service, an event fires, and ChatGPT responds with automation.

A concrete example from the keynote: ask ChatGPT to watch a project board for new tasks. When one arrives, ChatGPT reads the linked documents and drafts a plan, without you needing to be present. This is the building block for workflows that run in the background rather than requiring a conversation to trigger them.

Meetings plugin: notes and action items into Space

The Meetings plugin is a straightforward addition. It attends your meetings, takes notes, and saves personalised summaries and action items directly into ChatGPT Space. It launched in beta on macOS for Pro and Business users on 29 September 2026.

The integration with Space is the key detail. Notes do not sit in a separate tool; they land where the rest of your team’s context already lives, making them available to agents as well as teammates.

GPT-6.1 Sol: the model underneath everything

GPT-6.1 Sol is the new default model for agentic coding, computer use, and professional tasks. OpenAI positions it as near-Astra intelligence at one fifth of Astra’s standard token price. It is available across API, Plus, Pro, Business, Enterprise, and Edu tiers. Most of what shipped at DevDay runs on it.

The bigger picture

The thread connecting all of these announcements is that ChatGPT is being repositioned as a place where work happens rather than a tool you consult. Space and Pages give teams a shared surface. Plugin Extensions bring third-party tools into that surface. Cloud Codex keeps agents working after you step away. MCP Events means agents can respond to the world rather than waiting to be asked.

For enterprises, the value is in removing the gap between AI output and team workflow. For developers, the surface area available to build on has expanded considerably. For individual subscribers, the practical benefit is a workspace that starts to handle more of the coordination work that currently lives in your head.