OpenAI DevDay 2026: 20+ Launches and Business Takeaways

OpenAI's September 29 DevDay 2026 announcement is not one product launch. Its official recap groups more than 20 updates across personal agents, GPT-6.1 Sol, Codex, developer APIs, ChatGPT plugins and team collaboration. The clearest shift is from asking AI for an answer to assigning bounded work that can continue across tools—with permissions, review and deployment choices becoming as important as model capability.

For businesses, the immediate question is not which new feature to buy. It is which workflow is ready for an agent, what access that agent actually needs, and how a person will verify the result. Several announcements are available now, but others are limited previews or future releases. Those distinctions matter before committing to a roadmap.

What did OpenAI announce at DevDay 2026?

OpenAI's recap groups the releases into five connected areas:

• Personal and organizational agents: dots, specialist dots, ChatGPT Space and team tasks.
• Models and developer infrastructure: GPT-6.1 Sol, Ultrafast, the Decisions API, Agents API computer use, and Bedrock Managed Agents.
• Codex: cloud access, a refreshed CLI, desktop code review and Codex Security Cloud.
• ChatGPT extensibility: plugin extensions, improved plugin creation and discovery, plugins in Sites, and MCP events for automations.
• Work and account features: Pages, collaborative slides, ChatGPT in Slack and Teams, a Meetings plugin, shareable profiles, Sign in with ChatGPT, a new Pro tier and the OpenAI Marketplace.

This is a summary of OpenAI's published announcements, not a claim that all features are generally available in every country, workspace or plan.

Dots: always-on agents, with explicit boundaries

Dots are OpenAI's new personal agents, powered by GPT-6 Astra. OpenAI says a dot has its own cloud computer and browser, can use connected apps, and can continue work across multiple tasks. Users can contact a dot through ChatGPT and, where supported, Slack or Teams. The company describes a separate future path for specialist dots with organizational identities and access to company systems.

The control model is as important as the task model. OpenAI says users can inspect a dot's work, choose app access, set custom rules for actions, and review consequential steps. Its proactive background research is restricted to read-only tools; it cannot use that mode to send messages or change connected app content. OpenAI also says dots can make mistakes and that consequential work should be reviewed.

Dots are rolling out to Pro and Business Premium users in eligible markets. Enterprise, Edu and Healthcare workspaces can try a beta when an administrator enables it; the recap says this is off by default. Specialist dots are focused enterprise pilots, not a broadly available feature that a business should assume it can turn on today.

GPT-6.1 Sol: stronger workhorse model, but benchmark claims need context

GPT-6.1 Sol upgrades GPT-6 Sol for agentic coding, computer use and professional work. OpenAI positions it near GPT-6 Astra on selected evaluations at one-fifth of Astra's standard input and output token prices. The company reports gains on software engineering, PDF question-answering, multi-step automation and computer-use tests.

Those are OpenAI-reported evaluation results, not a guarantee that a company's own workflow will be five times cheaper or equally accurate. Real cost depends on prompts, output volume, tool calls, retries, caching and review. Teams should benchmark on their own tasks and compare total cost per accepted outcome, not just a model price or leaderboard score. OpenAI says GPT-6.1 Sol is available to API, Plus, Pro, Business, Enterprise and Edu users; plan and regional details should be confirmed before rollout.

The related Ultrafast tier is aimed at latency-sensitive work. GPT-6 Astra Ultrafast is available in specified API and higher-tier ChatGPT Work/Codex contexts, while GPT-6.1 Sol Ultrafast is coming soon. These should not be described as the same availability state.

New ways to build agents and business workflows

Several developer announcements are directly relevant to operational software:

• Decisions API is designed for finite, predefined answers to developer-defined questions, such as classifying content or routing a request. It is in limited preview, with a broader release planned. This is a narrow decision interface, not an endorsement of unsupervised decisions in high-stakes workflows.
• Agents API with computer use lets developers build agents that interact with software. The recap also describes multi-agent capabilities, tool search, tool calling and context compaction. Access is specified for the API and certain higher-tier Codex and ChatGPT Work plans.
• Bedrock Managed Agents is an OpenAI and Amazon collaboration for running OpenAI-powered agents within AWS and integrating with AWS resources. Suitability depends on an organization's architecture and governance needs.
• Private Intelligence includes Zero Data Retention with Private Safety Processing, while a Private Inference preview is planned for the fall. The preview is not yet a generally available control to assume in a procurement decision.

The practical lesson is to separate a model's ability to suggest a next step from authority to take it. Define data access, allowed actions, approvals, logging and rollback before giving an agent a production workflow.

What changed in Codex and ChatGPT collaboration?

OpenAI announced broader cloud access for Codex, a CLI update with voice steering and an /agents view, a desktop code-review experience, and Codex Security Cloud for repository scanning and proposed fixes. These are different tools for different stages of development; generated code and security findings still require an engineering review and tests before release.

For teams, ChatGPT Space offers shared context for people, ChatGPT and dots. Pages are collaborative documents that can support writing, research and other work inside ChatGPT. OpenAI also announced team tasks for recurring work, @ChatGPT in Slack and Microsoft Teams, and a Meetings plugin that can save summaries and action items to a Space. Collaborative slides were announced for the coming weeks, rather than as an already-available feature.

The shared-context story is useful, but permissions remain central: a team space, connector or meeting note should not automatically grant every agent access to every record or action.

Plugins, Sites and account changes

OpenAI is expanding plugins beyond conversational tool calls. The recap describes sidebar homes and interactive panels through plugin extensions, easier creation and discovery, supported plugins inside Sites, and support for proposed MCP events that can start automations when a connected app changes. User choice and access approval remain part of the plugin model.

Other announcements include shareable profiles, Sign in with ChatGPT across participating tools, a Pro 500 tier, and a marketplace for eligible enterprise customers. Each has different eligibility or rollout conditions. A team should verify the specific product documentation before assuming that a feature in the recap is included in its existing plan.

What should a business do next?

Start with one workflow rather than all 20-plus announcements. Write down its current steps, system owners, exception cases and measurable baseline. Then decide whether the job is best served by a personal agent, a custom agent, a finite decision endpoint, a plugin, or ordinary software automation.

A useful first pilot should answer five questions:

1. What exact task may the agent complete, and what is outside its scope?
2. Which apps and data are required, and which access can remain read-only?
3. Which actions need a person to approve them?
4. What evidence will prove that work was completed correctly?
5. What happens when the model, tool or integration fails?

Oplix helps businesses define those boundaries and build the surrounding AI agents, automation workflows and custom software. We can help scope a focused pilot, connect approved systems, design review paths and measure the result before expanding autonomy. If a real bottleneck is ready for that treatment, discuss an AI workflow with Oplix.

Source and timing note

The supplied OpenAI livestream page was a replay archive when checked; it did not itself contain the September 29 launch details. This article uses OpenAI's dated DevDay 2026 recap and direct product announcements. Availability described here is the launch status OpenAI reported on September 29, 2026 and may change.