Everything OpenAI Announced at DevDay 2026: Dots, GPT-6.1 Sol, Codex Cloud, Agents API, Plugins, Spaces & More

Everything OpenAI Announced at DevDay 2026: Dots, GPT-6.1 Sol, Codex Cloud, Agents API, Plugins, Spaces & More

OpenAI used DevDay 2026 in San Francisco to launch one of its broadest product updates yet. The September 29 event included more than 20 announcements spanning new AI agents, a new GPT model, faster inference, major Codex upgrades, new agent APIs, plugin infrastructure, collaborative workspaces, enterprise features, and a new ChatGPT plan.

The most important shift is bigger than any single feature: OpenAI is pushing beyond chat into persistent agents, shared workspaces, developer infrastructure, and software that can act across tools on a user’s behalf.

Here is a practical breakdown of what OpenAI actually announced, what is available now, and what matters most for developers and product teams.

DevDay 2026 at a Glance

• Dots: always-on agents powered by GPT-6 Astra.

• GPT-6.1 Sol: near-Astra capability at substantially lower API cost.

• Ultrafast: a premium high-speed inference tier.

• Private Intelligence: stronger privacy and confidential-computing options for enterprise AI.

• Codex in the cloud: reusable cloud development environments accessible across devices.

• Refreshed Codex CLI: voice control, an /agents view, improved worktrees, session resumption, and a cleaner terminal UI.

• Code Review: Codex-powered review workflows inside the ChatGPT desktop app.

• Codex Security Cloud: scheduled and continuous repository security analysis.

• Decisions API: structured real-time decisions over a finite set of possible answers.

• Agents API with computer use: managed agent infrastructure with computer interaction, multi-agent capabilities, tool search, tool calling, and context compaction.

• Bedrock Managed Agents powered by OpenAI: OpenAI agent capabilities integrated into AWS.

• Plugin extensions: richer native plugin experiences inside ChatGPT.

• Improved plugin creation and discovery: simpler creation, submission, ranking, and recommendations.

• Plugins in Sites: workspace Sites can host supported ChatGPT plugins.

• MCP Events: event-driven plugin automations based on the proposed MCP Events specification.

• ChatGPT Space: shared workspaces for teams, ChatGPT, and Dots.

• Pages: collaborative AI-native documents.

• Collaborative slides: real-time slide creation with people and agents.

• Teams and shared tasks: shared work plus scheduled or event-driven team automations.

• @ChatGPT in Slack and Microsoft Teams: ChatGPT directly inside workplace conversations.

• Meetings plugin: automatic notes, personalized summaries, and action items.

• Shareable profiles: a public or workspace-facing home for Sites, plugins, and reusable creations.

• Sign in with ChatGPT: ChatGPT identity and eligible plan usage across participating tools.

• Pro 500: a higher-usage ChatGPT plan with access to Ultrafast.

• OpenAI Marketplace: a way for eligible enterprise customers to apply OpenAI commitments toward approved partner software.

1. Dots: OpenAI’s Always-On Agents

The headline announcement was Dots, a new class of persistent agents designed to work continuously rather than wait for a single prompt.

OpenAI says Dots are powered by GPT-6 Astra, have their own cloud computer and browser, can learn from user feedback over time, and can connect to more than 4,000 apps through the plugin ecosystem.

A Dot can keep working toward a goal while the user is away, handle multiple projects, report progress, ask for decisions, and continue the same context across ChatGPT, Slack, and Microsoft Teams. OpenAI is also bringing voice calls to the experience.

The important architectural difference is persistence. Instead of a user repeatedly opening separate chats and manually restarting work, a Dot is intended to carry ongoing responsibility.

OpenAI’s examples include agents that watch customer feedback, prepare tested fixes and pull requests, update launch assets when requirements change, refresh research analyses as new data arrives, and prepare content from new source material.

Availability: OpenAI says Dots are rolling out to Pro and Business Premium users in eligible markets. Enterprise, Edu, and Healthcare workspaces can try the beta when an administrator enables it.

2. GPT-6.1 Sol: Near-Astra Capability at a Much Lower Price

OpenAI also introduced GPT-6.1 Sol, an upgrade to GPT-6 Sol aimed at coding, computer use, professional work, and agentic workflows.

The company positions GPT-6.1 Sol as offering performance close to GPT-6 Astra on several difficult workloads while costing roughly one-fifth as much on standard API input and output tokens.

OpenAI lists API pricing at:

• $2 per million input tokens

• $0.10 per million cached input tokens

• $10 per million output tokens

That cached-input price is especially relevant for long-running agents that repeatedly reuse system instructions, repository context, documents, or other stable context.

OpenAI reports meaningful gains over GPT-6 Sol across software engineering, professional document work, computer-use tasks, scientific workflows, and factual accuracy. GPT-6 Astra remains the higher-end option for the hardest tasks, but GPT-6.1 Sol creates a new cost-performance tier for production systems that do not need Astra on every request.

Availability: GPT-6.1 Sol is available through the API as gpt-6.1-sol and is available in ChatGPT Work and Codex for supported paid plans. OpenAI says it is not yet available in regular Chat.

3. Ultrafast: Premium Inference for Latency-Sensitive Workloads

Ultrafast is OpenAI’s new premium speed tier for workloads where latency matters more than inference cost.

According to OpenAI, GPT-6 Astra Ultrafast can reach up to 300 tokens per second in Codex, roughly an 8× speedup, and up to around 6× faster generation in the API.

This is particularly interesting for interactive coding, computer-use agents, voice-adjacent experiences, rapid tool loops, and workflows where waiting between model steps becomes the dominant UX problem.

GPT-6 Astra Ultrafast is available in the API and in ChatGPT Work and Codex on Pro 500 and Enterprise. GPT-6.1 Sol Ultrafast is planned next.

4. Private Intelligence

OpenAI grouped several enterprise privacy capabilities under Private Intelligence.

One component is Zero Data Retention with Private Safety Processing, which is designed to support automated safety review without giving OpenAI personnel access to the underlying customer content.

OpenAI also previewed Private Inference, planned for this fall, which combines confidential computing with verifiable controls.

For regulated and security-sensitive organizations, this is a significant direction: advanced model capability is being paired with infrastructure designed to reduce who or what can access sensitive inference data.

5. Codex in the Cloud

Codex can now run in reusable cloud development environments that are accessible across devices.

Instead of treating every remote coding task as an isolated sandbox, teams can use shared environments with approved configuration and permissions. A developer can start or inspect work from a computer, phone, or another device while Codex continues in the cloud.

This moves Codex closer to a persistent engineering runtime rather than a coding assistant tied to a single local terminal session.

6. A Refreshed Codex CLI

The Codex CLI received several workflow upgrades:

• voice-based task starting and steering;

• a new /agents view for delegating and tracking multiple tasks;

• improved prompt editing;

• better session resumption;

• improved worktree workflows;

• a cleaner terminal interface for long sessions.

The bigger theme is multi-agent supervision. As coding agents work on more tasks in parallel, the interface increasingly needs to help developers manage work rather than manually type every implementation step.

7. Codex Code Review

OpenAI introduced a new code-review experience in the ChatGPT desktop app.

Developers can inspect summaries and diffs, ask Codex questions about potential problems, and prepare feedback for GitHub pull requests or GitLab merge requests. Automatic review can also run in the cloud while the developer is away.

This extends Codex beyond code generation into another important part of the software lifecycle: understanding and validating changes before they merge.

8. Codex Security Cloud

Codex Security Cloud is a repository-security workflow for defenders and engineering teams.

It can scan GitHub repositories on demand, on a schedule, or as new commits arrive. OpenAI says Codex investigates findings, removes duplicates, and can prepare fixes in the cloud even when the user’s laptop is offline.

The product also includes access to models from OpenAI’s Daybreak Blue cybersecurity initiative without requiring a separate Daybreak application.

This is a notable expansion of coding agents from implementation into continuous security work.

9. Decisions API

The new Decisions API is designed for structured real-time decision making.

Instead of asking a general model to produce arbitrary free-form output, developers define a finite set of questions and allowed answers. Luna then evaluates supplied text or image context and returns decisions that an application can use directly.

Potential uses include:

• classification;

• content routing;

• workflow branching;

• choosing an agent’s next action;

• triage and prioritization.

This is a useful pattern when the application needs model intelligence but wants a tightly bounded output space.

Availability: limited preview at DevDay, with OpenAI saying a broader release is planned shortly afterward.

10. Agents API Now Supports Computer Use

OpenAI expanded the Agents API with computer use, allowing developers to build agents that interact with software directly.

The API also brings several capabilities associated with Codex into developer applications, including:

• multi-agent capabilities;

• tool search;

• tool calling;

• context compaction;

• managed underlying infrastructure.

This matters because production agent systems need much more than a model endpoint. They need execution, tools, context management, orchestration, and infrastructure that can survive long-running work.

11. Bedrock Managed Agents, Powered by OpenAI

OpenAI and Amazon also announced Bedrock Managed Agents powered by OpenAI.

The integration brings core OpenAI Agents API capabilities into AWS with customization and native connections to AWS resources, allowing organizations to build OpenAI-powered agents that run entirely within AWS infrastructure.

For companies already standardized on AWS, this reduces the distance between agent development and existing cloud governance, identity, infrastructure, and data systems.

12. Plugin Extensions

OpenAI is opening more of the platform it uses to build ChatGPT features so third-party developers can create richer experiences directly inside ChatGPT.

Plugin extensions can give a plugin a home in the sidebar, render interactive panels next to the conversation, and provide custom viewers for file types supported by the developer’s product.

This turns plugins from simple tool connections into much more complete product surfaces.

13. Easier Plugin Creation, Submission, and Discovery

OpenAI also announced improvements across the plugin lifecycle.

Plugin Creator is intended to make creation easier, while a redesigned submission flow provides clearer feedback. OpenAI is also improving directory ranking and in-conversation recommendations so relevant plugins can be discovered when they are useful.

Users still choose which plugins to connect and approve the access each plugin receives.

14. Sites Can Host Plugins

Supported ChatGPT plugins can now be added to Sites.

That means a workspace can create a shared Site and allow teammates to use the same application surface while still operating with their own connected accounts, permissions, and data.

OpenAI is also making automations easier to add and manage inside Sites so shared information can stay current.

15. MCP Events for Plugin Automations

OpenAI announced support for the proposed MCP Events specification.

This is important because it allows connected applications to trigger agent work when something happens, rather than requiring ChatGPT to poll repeatedly or wait for a person to initiate every task.

For example, a new task on a project board could trigger an automation that reads linked documents and prepares a plan in the background.

Event-driven execution is an important building block for always-on agents because it turns integrations into reactive systems rather than static tool catalogs.

16. ChatGPT Space

ChatGPT Space is a new shared workspace where teammates, ChatGPT, and Dots can collaborate around common knowledge and ongoing work.

Spaces can keep information organized according to instructions and provide a persistent home for team work that would otherwise be scattered across chats, documents, tasks, and connected tools.

It is available across Pro, Business, and Enterprise on desktop and web, with reading and sharing functionality also available on mobile and broader mobile editing planned.

17. Pages

Pages are AI-native collaborative documents inside ChatGPT.

A Page can support writing, research, charts, image generation, and information visualization while allowing teammates and AI to work on the same artifact.

The key idea is that the document is no longer separate from the AI workflow. The artifact itself becomes a shared surface for people and agents.

18. Collaborative Slides

OpenAI previewed collaborative slide creation inside ChatGPT.

Users will be able to create a deck from a conversation or template, let multiple teammates and agents edit it simultaneously, leave comments, present inside ChatGPT, and export to PowerPoint or Google Slides while preserving formatting.

OpenAI says this capability is coming to Pro, Business, and Enterprise in the coming weeks.

19. Teams and Shared Tasks

Business and Enterprise users can now create teams for sharing Pages, slides, plugins, spreadsheets, and recurring work.

Team tasks can run on schedules or in response to events such as new email or Slack messages. Connected tools can be used to gather information and take action, while teammates can collaboratively refine the instructions over time.

This combines collaboration, automation, and agents inside the same workspace.

20. @ChatGPT in Slack and Microsoft Teams

ChatGPT can now participate directly in workplace conversations.

Teams can mention @ChatGPT in a channel, thread, or direct message. It can work with admin-approved tools or the user’s own connected tools and permissions, and multiple people can add context or refine the result in the same conversation.

OpenAI says teammates can participate without each person needing a separate individual ChatGPT license under the supported Business and Enterprise setup.

21. Meetings Plugin

The new Meetings plugin can take meeting notes and save personalized summaries and action items into ChatGPT Space.

Those notes can remain private or be shared with a team, and ChatGPT can use them to update project plans or prepare follow-up work.

OpenAI says meeting audio is deleted after notes are generated and cannot be accessed or replayed afterward.

The feature is in beta in the ChatGPT desktop app on macOS for Pro and Business users, with Enterprise support planned.

22. Shareable Profiles

Shareable profiles give creators and teams a place to showcase Sites and plugins that other people can discover and reuse.

Within workspaces, teammates can also discover shared skills, while workspace-only skills remain limited to that organization.

This adds a social and discovery layer around reusable ChatGPT experiences.

23. Sign in with ChatGPT

OpenAI is expanding ChatGPT from a product account into an identity and usage layer for other tools.

Sign in with ChatGPT lets users connect supported tools with their ChatGPT account more easily. OpenAI also says eligible Plus and Pro users can use part of their ChatGPT plan allowance across participating partner products.

At launch, OpenAI named 16 partners, including tools such as Devin, Notion, Vercel, T3, OpenClaw, and Dactyl.

This is strategically important because it lets external products participate in the ChatGPT ecosystem without forcing every user to separately provision model billing in each tool.

24. A New Pro 500 Plan

OpenAI launched Pro 500, a higher-usage ChatGPT tier aimed at users who need significantly more capacity and speed.

OpenAI says Pro 500 provides 25× the ChatGPT Plus usage allowance and includes access to Ultrafast.

This plan also appears in several DevDay availability notes for the most compute-intensive capabilities.

25. OpenAI Marketplace

The OpenAI Marketplace is aimed at enterprise procurement.

Eligible enterprise customers can apply part of their existing OpenAI commitment toward approved partner software. OpenAI announced an initial set of 32 partners spanning design, customer experience, legal, cybersecurity, and AI infrastructure.

The first group includes companies such as Figma, Adobe, Sierra, Decagon, HubSpot, Salesforce, ServiceNow, Harvey, Legora, Palo Alto Networks, CrowdStrike, and Baseten.

What DevDay 2026 Means for Developers

The announcements can look like a long feature list, but they point toward a few clear platform-level changes.

1. Agents are becoming persistent systems, not single prompts

Dots, MCP Events, shared tasks, cloud Codex, and the Agents API all move in the same direction: AI work can continue after the initiating chat message ends.

That means agent architecture increasingly needs durable state, permissions, event handling, observability, approval boundaries, and long-running execution.

2. ChatGPT is becoming an application surface

Plugin extensions, Sites, Pages, Spaces, profiles, and richer discovery mean developers can build experiences that live inside ChatGPT rather than treating it only as a model client.

The combination of native UI, tools, user identity, plugins, connected data, and agent execution starts to look more like a platform than a chatbot integration layer.

3. Codex is expanding across the software lifecycle

Cloud environments, multi-agent task supervision, code review, and repository security move Codex beyond code generation.

The emerging workflow is closer to:

understand repository → plan work → implement → test → review → inspect security → prepare change → continue in the cloud

4. Cost and latency are becoming explicit model-routing dimensions

GPT-6.1 Sol and Ultrafast address two different production constraints.

GPT-6.1 Sol targets better economics for capable workloads. Ultrafast targets situations where latency is worth paying more for.

Production systems can increasingly route requests based on the combination of intelligence, price, and speed rather than choosing one model for everything.

5. MCP is moving toward event-driven automation

MCP Events is one of the quieter but technically important announcements.

Traditional tool calling is request-driven: the model decides to call a tool during a run. Event-driven integration adds the reverse direction: an external system can cause agent work to begin.

That is exactly the kind of primitive required for agents that monitor operational systems, project boards, repositories, inboxes, customer feedback, or business workflows continuously.

The Bottom Line

DevDay 2026 was not centered on one new model.

OpenAI launched a broader stack for ongoing AI work:

GPT models ↓ Agents API + computer use ↓ tools + plugins + MCP events ↓ persistent Dots and Codex workers ↓ Spaces, Pages, Sites and team tasks ↓ Slack, Teams and external partner products

GPT-6.1 Sol makes capable agent workloads cheaper. Ultrafast attacks latency. Codex is becoming a cloud engineering system. Dots introduce persistent personal agents. Plugins are becoming richer product surfaces. Spaces and Pages bring people and agents into shared artifacts. MCP Events adds an event-driven foundation for automations.

For developers, the most important takeaway is that the unit of AI software is shifting from “send a prompt and receive an answer” toward “give a persistent system a goal, tools, permissions, context, and events, then supervise the work it performs over time.”

Official Sources

• OpenAI — DevDay 2026 Recap

• OpenAI — Introducing Dots

• OpenAI — Introducing GPT-6.1 Sol

• OpenAI DevDay 2026

Availability and plan details in this article reflect OpenAI’s announcements on September 29, 2026 and may change as features roll out.