# OpenAI DevDay 2026: 25 Biggest Announcements and What They Mean

OpenAI DevDay 2026: 25 key launches including Dots, GPT-6.1 Sol, Codex Cloud, Decisions API, MCP Events, ChatGPT Space and Pro 500.

Canonical URL: https://aiidelist.com/blog/openai-devday-2026-announcements

Language: en

Published: 2026-09-30

Updated: 2026-09-30

## Key Takeaways

OpenAI DevDay 2026, held on September 29 in San Francisco, delivered more than 20 major announcements across ChatGPT, Codex, APIs, models, plugins, collaboration, and enterprise infrastructure.

The biggest shift was not a single model launch. OpenAI is moving from prompt-and-response software toward **persistent AI workers** that can keep context, use tools, react to events, work in the cloud, collaborate with teams, and continue tasks while the user is away.

The most important announcements include:

- **Dots**, OpenAI's new always-on agents powered by GPT-6 Astra.
- **GPT-6.1 Sol**, positioned near Astra for coding, computer use, and professional work at much lower token prices.
- **Ultrafast**, a premium inference tier for latency-sensitive agent workflows.
- **Decisions API** and **Agents API with Computer Use**.
- Major **Codex Cloud, Code Review, CLI, and security** upgrades.
- A broader **ChatGPT plugin platform**, including interactive extensions and MCP Events.
- **ChatGPT Space, Pages, collaborative slides, teams, and meetings**.
- **Sign in with ChatGPT**, **Pro 500**, and the **OpenAI Marketplace**.

## OpenAI DevDay 2026 at a Glance

OpenAI DevDay 2026 took place on **September 29, 2026 at Fort Mason in San Francisco**. The event focused on APIs, developer tooling, cloud agents, coding, collaboration, and OpenAI's expanding application ecosystem.

The announcements can be grouped into five broader themes:

1. **Persistent agents** that keep working after a prompt ends.
2. **Cheaper and faster frontier models** for production agent workloads.
3. **Cloud-native software engineering** through Codex.
4. **ChatGPT as an application and collaboration platform**.
5. **A broader commercial ecosystem** around identity, subscriptions, and enterprise software.

## 1. Dots: OpenAI's Always-On Agents

Dots are OpenAI's clearest attempt yet to move beyond the conventional chatbot model.

A dot is an **always-on agent powered by GPT-6 Astra** with its own cloud computer, browser, persistent working context, and access to connected apps.

A developer-focused dot could:

- Monitor customer feedback.
- Detect recurring bugs or feature requests.
- Scope a fix.
- Modify and test code.
- Prepare a pull request.
- Attach a video showing what changed.
- Ask for approval before consequential actions.

The architectural shift is:

`prompt -> response`

becoming:

`goal -> monitor -> reason -> work -> request approval -> continue`

The defining feature is persistence rather than a single agent run.

## 2. GPT-6.1 Sol Brings Near-Astra Capability Downmarket

GPT-6.1 Sol is the major model release of DevDay 2026.

OpenAI positions it as especially strong in **agentic coding, computer use, and professional work**, while targeting much lower cost than GPT-6 Astra.

Published API pricing includes:

- **Input:** $2.00 per 1M tokens
- **Cached input:** $0.10 per 1M tokens
- **Output:** $10.00 per 1M tokens
- **Context window:** 1,050,000 tokens
- **Maximum output:** 128,000 tokens

The model ID is `gpt-6.1-sol`.

For production systems, this makes model routing more important. Many workloads that previously required the most expensive model may now be economical on Sol.

## 3. Ultrafast Turns Latency Into a Product Tier

OpenAI introduced **Ultrafast** as a premium inference mode for latency-sensitive workloads.

The company says it can deliver substantially faster generation in Codex and the API.

This creates a more complex model-selection matrix:

- **Model intelligence**
- **Reasoning effort**
- **Token cost**
- **Latency tier**

For interactive coding, browser control, and rapid tool-calling loops, speed can matter almost as much as benchmark performance.

## 4. Private Intelligence Targets Sensitive Enterprise Workloads

OpenAI introduced **Private Intelligence** for organizations handling sensitive data.

The initiative includes privacy-focused processing designed for cases where enterprises need AI capability without unnecessarily exposing internal content.

This is particularly relevant for:

- Financial services
- Legal workflows
- Healthcare
- Security operations
- Proprietary research
- Internal enterprise knowledge

Enterprise AI competition is increasingly about both model capability and data-handling guarantees.

## 5. Codex in the Cloud

Codex can now run in reusable cloud development environments rather than depending entirely on a developer's local machine.

Teams can configure shared environments with approved:

- Repositories
- Dependencies
- Tools
- Settings
- Permissions

A task can continue remotely even after the developer closes a laptop.

This moves Codex closer to a persistent software-engineering worker rather than a terminal-only assistant.

## 6. Codex CLI Gets Better Multi-Agent Control

The refreshed Codex CLI adds more tools for delegating and monitoring parallel coding tasks.

Improvements include:

- Voice-driven task initiation
- Better agent views
- Session resumption
- Git worktree support
- Improved terminal readability
- Multi-task oversight

For developers running several agents against one repository, visibility and task isolation are becoming core features.

## 7. Code Review Moves Into ChatGPT

OpenAI expanded Codex into code review workflows.

Developers can:

- Read change summaries.
- Explore diffs.
- Ask Codex about suspicious code.
- Prepare GitHub pull request feedback.
- Work with GitLab merge requests.
- Run an initial cloud review automatically.

The key shift is from **generating code** to participating in the complete software-development loop.

## 8. Codex Security Cloud Adds Repository Scanning

Codex Security Cloud extends Codex into security workflows.

It can scan repositories, inspect commits, investigate findings, deduplicate issues, and prepare fixes.

The Codex product surface is therefore expanding from:

`write code -> review code -> secure code`

This is strategically important because security budgets can be substantially larger than individual developer-tool budgets.

## 9. Decisions API Creates a Dedicated AI Decision Layer

The **Decisions API** is one of the most important developer announcements from DevDay 2026.

Instead of asking a model to produce arbitrary text, developers define:

- A question
- A finite set of allowed answers
- Context supplied as text or images

The model then selects an answer.

Typical uses include:

- Classification
- Request routing
- Policy selection
- Workflow branching
- Choosing an agent's next action

Conceptually:

```text
Context
  ↓
Decision model
  ↓
A | B | C | D
  ↓
Deterministic workflow
```

This matters because many production AI calls do not require prose. They require a reliable decision that another system can execute.

## 10. Agents API Adds Computer Use

The Agents API now supports **computer use**, allowing agents to interact with software interfaces.

It also expands OpenAI's managed agent infrastructure with capabilities such as:

- Multi-agent orchestration
- Tool search
- Tool calling
- Context compaction
- Managed execution

The broader direction is clear: OpenAI wants the API to become an agent runtime, not only a model endpoint.

## 11. Amazon Bedrock Managed Agents Bring OpenAI Agents Into AWS

OpenAI and Amazon introduced managed agent capabilities for AWS environments.

For enterprises already standardized on AWS, this reduces the need to choose between OpenAI's agent stack and existing cloud governance.

This matters most for organizations that need:

- AWS-native permissions
- Existing cloud resources
- Centralized governance
- Enterprise deployment controls

## 12. Plugin Extensions Turn ChatGPT Plugins Into Interfaces

Plugin extensions significantly broaden what a ChatGPT plugin can be.

Developers can build experiences such as:

- Sidebar destinations
- Interactive panels
- File viewers
- Interfaces that remain visible beside a conversation

Plugins are therefore moving from simple tool integrations toward **applications living inside ChatGPT**.

## 13. Plugin Creator and Discovery Improve

OpenAI is also reducing friction across the plugin lifecycle.

DevDay highlighted:

- **Plugin Creator**
- Redesigned submission flows
- Better submission feedback
- Improved ranking
- Contextual recommendations

As plugin supply grows, discovery may become as important as the integration itself.

## 14. ChatGPT Sites Can Host Plugins

Supported plugins can be added to ChatGPT Sites.

This creates a path from an internal AI page or mini-app to a connected operational tool without forcing teams to build a traditional standalone SaaS frontend.

For internal workflows, this could reduce both development time and user-training costs.

## 15. MCP Events Make Agent Workflows Event-Driven

OpenAI announced support for **MCP Events**.

Traditional tool use is request-driven:

`user asks -> model calls tool`

MCP Events enables:

`external event -> agent wakes up -> reads context -> acts`

For example, a new project-board task could trigger an agent to read linked documents and prepare a plan automatically.

This moves ChatGPT closer to event-driven automation platforms.

## 16. ChatGPT Space Becomes a Shared AI Workspace

**ChatGPT Space** is a collaborative environment where teammates, ChatGPT, and agents can work from shared knowledge.

Instead of organizing work around disconnected one-to-one chats, a Space can become the persistent context for a project or team.

The strategic direction is clear: ChatGPT is moving from personal assistant to team operating environment.

## 17. Pages Bring AI-Native Documents Into ChatGPT

**Pages** are documents designed for collaboration between humans and agents.

Inside a Page, users can:

- Write
- Research
- Generate charts
- Create images
- Visualize information
- Invite teammates to comment

The document itself becomes an active AI working surface rather than a static artifact.

## 18. Collaborative Slides Are Coming

OpenAI announced collaborative slide creation.

Multiple teammates and agents will be able to edit a presentation together, comment, present, and export to common presentation formats.

The important detail is the workflow: people and agents can work on the same deck rather than generating a one-off presentation.

## 19. Teams and Shared Tasks Add Recurring Work

Business and Enterprise users can create teams that share work artifacts and connected tools.

They can also delegate recurring work through team tasks.

Tasks can potentially run:

- On a schedule
- When a new email arrives
- When a Slack message appears
- In response to supported events

This connects collaboration, plugins, agents, and automation into one product layer.

## 20. @ChatGPT Arrives in Slack and Microsoft Teams

Business users can invoke ChatGPT directly inside Slack or Microsoft Teams.

That matters because OpenAI no longer needs every workflow to start on the ChatGPT website.

ChatGPT can become part of:

- Team channels
- Threads
- Direct messages
- Existing company workflows

This is an important enterprise distribution strategy.

## 21. Meetings Plugin Creates Notes and Action Items

The **Meetings plugin** turns meeting audio into summaries and action items.

Those notes can then feed downstream work such as:

- Updating project plans
- Drafting follow-ups
- Assigning action items
- Sharing decisions

This turns meeting capture into an input for persistent agent workflows.

## 22. Shareable Profiles Create a Builder Discovery Layer

Shareable profiles allow users to showcase the Sites and plugins they have built.

Although smaller than Dots or GPT-6.1 Sol, this adds an identity and discovery layer to ChatGPT.

If plugin discovery becomes meaningful, public profiles could function like developer storefronts or portfolios.

## 23. Sign in with ChatGPT Expands Identity and Compute Portability

**Sign in with ChatGPT** lets users authenticate with supported third-party applications through their ChatGPT identity.

The more important idea is usage portability: participating tools can potentially rely on a user's existing ChatGPT plan rather than forcing every user to manage a separate API key.

This begins to make a ChatGPT subscription resemble an **AI identity plus compute wallet**.

## 24. Pro 500 Adds a High-Usage Individual Tier

OpenAI introduced **Pro 500**, a higher-usage personal plan.

The strategic signal is more important than the plan name.

Frontier AI subscriptions are increasingly segmented by:

- Usage allowance
- Concurrency
- Agent work
- Cloud execution
- Speed
- Premium inference tiers

The product being sold is increasingly **compute capacity and autonomous work**, not merely chat messages.

## 25. OpenAI Marketplace Adds Enterprise Software Distribution

The **OpenAI Marketplace** lets eligible enterprise customers purchase approved partner software through OpenAI-linked commercial arrangements.

This gives OpenAI several roles at once:

- AI infrastructure provider
- Identity provider
- Agent platform
- Application distribution channel
- Enterprise software marketplace

For startups building on OpenAI, marketplace distribution could eventually become an important enterprise go-to-market route.

## Which DevDay 2026 Announcements Matter Most for Developers?

The 25 announcements are not equally important for every developer.

Five deserve especially close attention.

### GPT-6.1 Sol

Lower pricing and strong agentic performance can materially change the economics of production AI.

The **$0.10 per million cached input tokens** rate is particularly relevant for systems that repeatedly reuse:

- Repository context
- Long system prompts
- Agent memory
- Product documentation
- Customer-specific context

### Decisions API

A constrained decision endpoint can replace full generative calls when an application only needs a classification or next step.

That can improve control, reduce unnecessary output, and simplify downstream automation.

### Agents API with Computer Use

Developers can increasingly outsource parts of the agent runtime to OpenAI rather than maintaining their own orchestration, context management, and computer-control stack.

### MCP Events

MCP Events changes the trigger model from **user asks AI** to **software event activates AI**.

That is a foundational requirement for background agents.

### Plugin Extensions

Interactive plugin interfaces turn ChatGPT into a distribution surface where third-party applications can live directly beside a conversation.

## The Bigger Shift: From Chatbots to Persistent AI Workers

The individual announcements make more sense when viewed as one architecture.

A modern OpenAI-based worker can potentially have:

- **Intelligence:** GPT-6.1 Sol or GPT-6 Astra
- **Persistent execution:** Dots or cloud agents
- **A computer:** Computer Use
- **Tools:** Plugins and MCP
- **Triggers:** MCP Events and team tasks
- **Shared context:** Space
- **Work artifacts:** Pages, slides, spreadsheets, and code
- **Communication channels:** ChatGPT, Slack, Teams, and voice
- **Identity:** Sign in with ChatGPT
- **Distribution:** Plugin discovery and profiles
- **Commercial distribution:** OpenAI Marketplace

That looks much more like an operating layer for AI workers than a conventional chatbot.

## What DevDay 2026 Means for AI Startups

DevDay creates both opportunities and risks.

### Opportunity: Build on the Distribution Layer

Startups no longer need to think only in terms of:

`website -> signup -> API key -> SaaS dashboard`

A second distribution architecture is emerging:

`ChatGPT -> plugin or extension -> authentication -> connected workflow`

### Risk: Thin Agent Wrappers Face More Platform Pressure

Products whose primary value is simply:

- Running prompts in the background
- Calling common SaaS tools
- Managing basic agent loops
- Reviewing straightforward code
- Summarizing meetings
- Generating standard documents

now overlap more directly with OpenAI's first-party stack.

Durable differentiation is more likely to come from:

- Proprietary data
- Deep vertical workflows
- Domain-specific evaluation
- Unique distribution
- Compliance
- Human networks
- Specialized interfaces
- Workflow ownership

### Opportunity: Event-Driven Vertical Agents

The combination of MCP Events, Decisions API, Agents API, plugins, and persistent agents makes event-driven vertical workflows easier to build.

Examples include:

- A support agent that reacts to escalated tickets.
- A finance agent that reviews incoming invoices.
- A development agent that investigates new bug reports.
- A sales agent that updates proposals when requirements change.
- A research agent that reruns analysis when fresh data arrives.

The opportunity is no longer simply to create a better chat window. It is to own a recurring business process.

## Common Mistakes When Evaluating DevDay 2026

### Mistake 1: Treating Dots as Just Another ChatGPT Agent

The defining feature is persistence.

Dots are designed to maintain goals, context, tools, and ongoing responsibilities rather than execute one prompted task.

### Mistake 2: Choosing GPT-6 Astra for Every Hard Task

GPT-6.1 Sol exists to make capable agentic work more economical.

Production systems should benchmark both models and route only the hardest or highest-value tasks to Astra.

### Mistake 3: Ignoring Cached Input Economics

For context-heavy agents, cache pricing can materially change total cost.

Architecture choices around reusable prompts, stable context prefixes, and memory therefore matter.

### Mistake 4: Treating Ultrafast as a Universal Upgrade

Speed matters when latency changes the user experience or allows more tool iterations.

For asynchronous jobs or background research, paying a premium for maximum speed may provide little business value.

### Mistake 5: Building Plugins as Simple Tool Wrappers

Plugin Extensions, interactive panels, profiles, recommendations, and Sites mean successful ChatGPT integrations increasingly require product design, not only an MCP server.

## What to Watch After DevDay 2026

Several announcements are still rolling out rather than universally available.

Important follow-up areas include:

- Broader **Dots** availability.
- Wider release of **Decisions API**.
- **GPT-6.1 Sol Ultrafast**.
- **Collaborative Slides**.
- Further development of **MCP Events**.
- More **Sign in with ChatGPT** partners.
- Expansion of the **OpenAI Marketplace**.
- Changes in plugin ranking and discovery.

Developers should also watch actual quotas and pricing closely. In agentic products, background execution, repeated tool calls, long contexts, concurrency, and premium speed can matter more than headline token prices.

## Conclusion

OpenAI DevDay 2026 was not primarily about one new model. It was about assembling the pieces of a much larger platform.

**Dots provide persistence. GPT-6.1 Sol lowers the cost of capable agentic work. Codex moves software engineering into the cloud. Decisions API and Agents API provide agent infrastructure. MCP Events add triggers. Plugins provide tools and interfaces. Space and Pages provide shared context and artifacts. Sign in with ChatGPT adds identity, while Marketplace adds commercial distribution.**

The result is a clear direction: OpenAI wants ChatGPT to become a place where **people, applications, and persistent AI workers operate together**, rather than simply a website where users ask a model questions.

For developers, the practical next question is simple: **which parts of an existing workflow should remain user-triggered, and which can now become persistent, event-driven agent processes?**