AIJiten

日本語

OpenAI DevDay 2026 Recap: What Dots, GPT-6.1 Sol, and Space Can Do

Articles
A brass robotic arm arranging paper cards on a wooden workbench, with a person's hand reviewing them.

OpenAI made more than 20 major announcements at DevDay 2026, held on September 29, 2026 in the United States—September 30 in Japan.

The announcements include Dots, which keep working after a conversation ends; the new GPT-6.1 Sol model; and Pages, where people and AI can edit documents together. They expand what you can delegate when preparing documents, researching a topic, or writing code.

Supported plans and devices vary by feature. Some features are rolling out gradually, while others are planned for a later release.

Dots keep working after the conversation ends

Dots are always-on AI agents powered by GPT-6 Astra. An AI agent works toward a requested goal by planning the steps it needs and using tools to carry them out.

Each dot has its own cloud computer and web browser. It can research topics, prepare documents, analyze data, and develop software. It continues working between conversations and contacts you when a decision needs your input.

Cloud work can continue even when you turn off your computer. However, if you connect your own PC so that a dot can use its files and apps, that computer must remain online with the ChatGPT app open.

You can give instructions and check progress through ChatGPT, Slack, or Microsoft Teams. For example, you could ask a dot to prepare a draft post from an interview transcript or update project documents as the project changes.

Dots are rolling out to eligible plans

  • Pro plans (Pro 100, Pro 200, and Pro 500): The official documentation specifies users who are “over 18.” Dots are rolling out in regions outside the European Economic Area (EEA), the United Kingdom, and Switzerland. Japan is not among the excluded regions, but having an eligible subscription does not guarantee that the feature will appear immediately.
  • Business Premium: Dots are rolling out worldwide. This availability differs from the standard Business plan.
  • Enterprise: A workspace administrator must explicitly enable Dots in settings. They are off by default.
  • Edu and Healthcare: Dots are available as a beta that a workspace administrator must enable.

The first dot is included at no additional charge on eligible Pro and Business Premium plans. Conversations with your dot do not count against your ChatGPT usage allowance. However, ChatGPT Work and Codex tasks that a dot starts or manages count against the allowances for those products.

You can choose which apps to connect and which actions require approval. Proactive background research initiated by a dot is read-only: it cannot send messages or change documents. Work you request is subject to separate permission and approval rules. Check important deliverables yourself before using them.

An always-on assistant with uses similar to OpenClaw

Dots support workflows similar to those of the open-source OpenClaw project. Both let you ask an always-on AI to work through familiar messaging services and delegate ongoing tasks.

OpenClaw runs on a computer or server you manage. It connects messaging services such as Slack and Telegram to AI agents and supports conversation memory and scheduled runs. You choose the operating environment and AI model, then configure and maintain the system for your needs. Dots, by comparison, use GPT-6 Astra in a cloud environment provided by OpenAI and are available through eligible ChatGPT plans. Their uses are similar—asking an AI to handle ongoing work through chat—but the hosting environment and the way you operate them differ.

GPT-6.1 Sol arrives in Work, Codex, and the API

GPT-6.1 Sol is an update to GPT-6 Sol.

According to OpenAI's evaluations, it delivers performance close to GPT-6 Astra in coding, computer use, and professional work. You will still need to test it on your own tasks to see whether those results carry over.

For context on the preceding models, see our introduction to GPT-6 Sol and Luna.

Availability has begun in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu. At the time of the announcement, it was not yet available in the standard Chat mode.

API model ID and pricing

Developers can access the model through the API using the model ID gpt-6.1-sol. An API lets an external application or program call an AI model. Charges are based on the number of tokens—the units used to process text.

Pricing item (per one million tokens) Price (USD)
Standard input US$2.00
Cached input US$0.10
Standard output US$10.00

Standard input and output prices are one-fifth of Astra's. Cached input is priced separately and is not included in that comparison. The comparison also does not mean that a monthly subscription or the total cost of completing a task becomes one-fifth as expensive.

GPT-6.1 Sol Ultrafast is planned for release in the coming days. Its availability therefore differs from that of the standard Sol model.

Space and Pages let people and AI edit documents together

Space brings related pages, reference files, and shared tasks together in one place. Pages are the documents within that environment that people and ChatGPT can write and edit together.

For example, you can ask ChatGPT to draft a plan using a proposal and research data, then have teammates refine it through edits and comments.

  • Eligible plans: Pro, Business, and Enterprise
  • Supported environments: The desktop app and web browsers

On mobile, you can find, read, and share pages; creating and editing them is planned for a later release. Collaborative slides are also planned for release in the coming weeks.

Team automation and who can see shared information

Business and Enterprise also have Team Tasks for delegating recurring team work. OpenAI announced integrations that let you request work by mentioning @ChatGPT in Slack or Microsoft Teams. Administrator settings and the permissions of connected apps apply.

Before sharing, check who can access both the original files and the page. Permissions on linked source files remain in place. However, content copied or summarized into a page becomes visible to people who can access that page.

Codex gains cloud workflows and security scanning updates

The Codex updates are aimed at people who write code or manage applications.

Reuse prepared development environments with Codex Cloud

With Codex Cloud, you can prepare and reuse a repository—the place where source code is stored—along with the required software, tools, and access settings. Each new task starts in an independent workspace based on that prepared environment. You can make further adjustments for the task when needed.

Cloud tasks can continue after you close your computer. If you instead connect your own PC for remote operation, that computer must remain online.

  • Eligible plans: Plus, Pro, Business, Healthcare, Edu, and Enterprise
  • Requirements: Device requirements and organization administrator settings apply.

Codex Security Cloud scans code on GitHub

Codex Security Cloud is a research preview for investigating security weaknesses in software. It can scan an entire GitHub codebase or monitor commits—individual code changes—and prepare evidence and proposed fixes for issues it finds.

  • Eligible plans: Pro, Business, Enterprise, and Edu
  • Supported environments: Web browsers and the desktop app

A person should review the evidence and proposed fixes before applying them. A scan that finds no issues does not guarantee that the software is secure.

Other Codex announcements include a refreshed command-line interface (CLI) with voice input, an /agents view for checking the status of multiple AI agents, and code review features in the desktop app.

Astra Ultrafast and Pro 500 offer faster generation

GPT-6 Astra Ultrafast offers up to eight times the standard token generation speed in Codex. This comparison measures how quickly text or code is generated, rather than the time needed to complete an entire task, including network requests and external tool operations.

It is included in Pro 500, which costs US$500 per month, and is not included in Pro 100 or Pro 200. Eligible Enterprise and Edu plans also have access under specified conditions. It is off by default for Enterprise.

On Pro 500, Ultrafast consumes the usage allowance at eight times the standard rate. Consider that usage cost when deciding whether a task needs the faster response. It is a separate speed setting from Fast mode.

Decisions API makes choices in a way similar to Jev

The newly announced Decisions API uses Luna to return decisions based on questions and answer options defined by a developer. Given text or images, it can classify content, route an inquiry, or select an AI agent's next action.

Its uses are similar to those of Jev from TypeSafe AI. Jev uses its own System One model to return typed decisions and probabilities instead of freely generated text. For example, both can support a workflow that decides whether an incoming inquiry should go to a billing team or a technical support team.

They differ in their underlying models and the information they can accept. At the time of our review, Jev accepted text input only, while Decisions API also supported images. Decisions API was in limited preview at the time of the announcement, with broader availability planned in the coming days.

Our article What Is Jev? An AI That Returns Decisions Instead of Prose explains how Jev handles decisions and probabilities in more detail.

Plugins gain richer interfaces and external app integrations

Developer features for connecting ChatGPT to external tools and web services have also expanded.

Give a plugin its own workspace inside ChatGPT

Plugin Extensions let developers display an app in the sidebar or a panel beside a conversation, or create viewers and editors for supported files. Mentions that let users select content from the input field are desktop-only. Extensions for Free and Go on the web are planned for a later release. (Official documentation)

Start work when a new message or comment arrives

MCP Events let an event such as a new message or comment in a connected service trigger a specified task. MCP is a way to connect AI to external tools. This feature requires MCP 2.0 and a compatible plugin. ChatGPT supports a webhook approach based on the proposed specification, so it does not work with every existing plugin. (Official documentation)

Use your ChatGPT account and allowance in external apps

Sign in with ChatGPT lets you use your ChatGPT account to log in to external apps. Signing in and using an eligible Plus or Pro allowance for AI processing are separate capabilities. Integration in commercial apps is being tested with selected partners, and there is also a mechanism for using the allowance in open-source tools.

Eligible AI processing in an external app consumes your plan's allowance. Signing in alone does not grant access to your conversation history. (Official documentation)

OpenClaw is also listed in the official announcement as an integration that can use the ChatGPT allowance. (Official announcement)

Check what is available, rolling out, or still planned

The availability of major announced features varies by plan and region. The following table summarizes their status at the time of the announcement.

Feature Status at launch Check before use
Dots Gradual rollout; beta access for teams Pro age and region rules; Business Premium; admin settings
GPT-6.1 Sol Work, Codex, and API access has begun Not yet in standard Chat mode
GPT-6.1 Sol Ultrafast Planned for the coming days Release date differs from standard Sol
Space and Pages Pro, Business, and Enterprise Desktop and web; creating and editing on mobile come later
Shared slides Planned for the coming weeks Check access after release
Codex Security Cloud Research preview Pro, Business, Enterprise, or Edu; GitHub link
Astra Ultrafast On eligible plans Pro 500 and other eligible plans; check usage rates
Decisions API Limited preview; wider release in the coming days For developers; text or images with a fixed set of answer options

Choose one task to try

Start by checking whether a feature appears in your current plan. A feature with a gradual rollout, such as Dots, may not appear immediately even if you have an eligible subscription. An administrator may also need to enable it for a work account.

When a feature is available, choose one task whose result you can check yourself: edit a short plan together in Pages or ask GPT-6.1 Sol to make a small code change. When assigning ongoing work to a dot, specify the reference material it should use and which actions it should pause before taking until you approve them.

It also helps to distinguish between ChatGPT Work and Codex when choosing where to carry out a task. Our guide to Codex tasks, threads, and subagents explains the units you can use to organize work.

Review AI deliverables alongside your plan's allowance or API usage. That will help you decide which task to delegate next.

Sources

This article is based on official information available as of September 30, 2026. Availability, eligible plans, and prices may change.

Related articles

Search this site