TL;DR: OpenAI used its Sept. 29 DevDay in San Francisco to launch always-on Dots agents, the GPT-6.1 Sol model, and more than 20 developer and ChatGPT updates for its 1.2 billion weekly users.

OpenAI bets DevDay on agents that keep working when you stop typing

OpenAI turned its annual developer gathering on Tuesday, Sept. 29, 2026, in San Francisco into a showcase for autonomous software, headlined by always-on assistants called Dots, a lower-cost GPT-6.1 Sol model and a new $500-per-month performance tier. Chief Executive Sam Altman led the keynote, according to live coverage from CNBC and a roundup from The Verge.

The company framed DevDay 2026 as its largest yet, with more than 20 announcements spanning ChatGPT, Codex, models and new forms of collaboration, according to the DevDay 2026 Recap published by OpenAI. The same recap put collective ChatGPT reach at 1.2 billion weekly users, a figure also cited in reporting from The Verge.

The timing was tense. CNBC reported that the event followed disclosures of unintended model behavior and a decision to hold back the latest Astra model on safety grounds. Axios reported that OpenAI said it would not release its newest flagship model, GPT-6.1 Astra, because of security concerns. Altman addressed that withholding decision on stage, while Chief Financial Officer Sarah Friar told CNBC the company would pace frontier work when necessary.

What are Dots and who gets them first?

Dots are pitched as persistent helpers that learn preferences, stay active in the background and take ongoing responsibilities off a user's plate, according to the DevDay 2026 Recap from OpenAI. Users can hand them tasks and projects, link other applications and let work continue without constant prompting.

Axios described Dots as OpenAI's largest consumer bet at the show, powered by the highly capable Astra model and positioned as an "extension of you." The Verge characterized Dots as a rival to Meta's recently launched Muse assistant, but with a different business model at launch.

Unlike Muse, which The Verge noted is available for free, Dots will initially be available to paid subscribers of ChatGPT Pro, Business Premium and Enterprise, according to The Verge. Decrypt also reported availability for Pro and Business Premium users in eligible markets, with agents running on their own cloud computers.

Friar, speaking to CNBC at Fort Mason on Sept. 29, said the agents are intended to help people have a better day, be more productive and potentially finish earlier. She described the experience as an agent that can tap a user on the shoulder with reminders, according to CNBC's live coverage.

GPT-6.1 Sol brings near-Astra power at lower token prices

The second pillar was GPT-6.1 Sol, presented as a major upgrade to GPT-6 Sol with strong results for agentic coding, computer use and professional work, according to the DevDay 2026 Recap from OpenAI. OpenAI said the system delivers near-Astra intelligence at one-fifth of Astra's standard input and output token prices, giving developers room to run longer workflows on the same budget.

Axios summarized the same trade-off, reporting that Sol approaches Astra performance on coding and computer use at a sharply lower cost. The OpenAI developer community summary echoed that standard token prices are one-fifth of Astra's.

For speed-sensitive work, OpenAI introduced Ultrafast. GPT-6 Astra Ultrafast is available in the API and in ChatGPT Work and Codex on Pro 500 and Enterprise plans, with GPT-6.1 Sol Ultrafast coming soon, according to the DevDay 2026 Recap from OpenAI. A developer-community summary of the announcements reported up to 8-times faster token generation in Codex and 6-times faster generation in the API.

The recap also cited platform efficiency gains, including 45% lower API time to first token and more than 30% faster tool calls and workflows, according to the developer-community summary of DevDay materials.

Pro 500 and the $500 speed tier

OpenAI added a higher-priced subscription option around power and speed. The Verge reported a new ChatGPT Pro plan costing $500 per month. Decrypt described a $500 speed tier linked to the launch.

According to the developer-community summary, Pro 500 adds 25 times the Plus usage allowance plus Ultrafast access. The DevDay 2026 Recap from OpenAI ties Ultrafast in ChatGPT Work and Codex to Pro 500 and Enterprise plans.

Friar addressed pricing skepticism in an interview with CNBC. She recalled that when OpenAI launched its $200 offering, observers thought no one would pay that amount monthly. Demand proved stronger than expected, she said, with customers moving toward consumption-based credits because small businesses saw strong outcomes, according to CNBC.

Friar also told CNBC that OpenAI had an "incredible" third quarter, with 70% quarter-over-quarter growth and an enterprise business that has doubled since July. She attributed consumer momentum to better models, images, voice and seasonal back-to-school and back-to-work demand, according to CNBC. She declined to comment on early fundraising discussions and said the company, which raised $122 billion in March, remains very well capitalized, according to CNBC.

Codex moves into the cloud, terminal and code review

Coding assistant Codex received several of the more than 20 announcements.

Codex can now run on a computer, remotely from a phone or in the cloud from any device, with reusable development environments for fast starts and shared team settings and permissions, according to the DevDay 2026 Recap from OpenAI. Availability spans Plus, Pro, Business, Healthcare, Education and Enterprise. Axios noted the practical effect: developers can start or monitor coding work from a phone while tasks continue remotely.

The Codex command-line interface was refreshed with voice input to start and steer tasks, a new /agents view for delegating and tracking parallel work, plus improved prompt editing, session resumption, worktree support and a cleaner terminal interface for long sessions, according to the DevDay 2026 Recap from OpenAI.

Code review moved into the ChatGPT desktop app. Users can read summaries, explore diffs and question Codex about possible problems before commenting on GitHub pull requests or GitLab merge requests, according to the same OpenAI recap. Automatic reviews can make an initial pass in the cloud while a developer is away.

For defenders, Codex Security Cloud can scan entire GitHub repositories on demand or on a schedule, with continuing checks of new commits, according to the DevDay 2026 Recap from OpenAI. The system investigates findings, removes duplicates and prepares fixes in the cloud, even with a laptop closed, and includes access to models offered through Daybreak Blue without a separate Daybreak application.

Agents API, Decisions API and AWS partnership

For builders, the Agents API now supports computer use, allowing agents to operate software through its interface to finish tasks, according to the DevDay 2026 Recap from OpenAI. The update also brings Codex multi-agent capabilities, tool search, tool calling and context compaction into outside applications, with OpenAI operating the underlying infrastructure. It is available through the API and in Codex and ChatGPT Work on Pro 500 and Enterprise.

A separate Decisions API focuses the smaller Luna model on a defined set of user-specified questions with finite answers, according to the same OpenAI recap. Developers provide context as text or images and receive answers for classifying content, routing requests or selecting an agent's next step. It was available in limited preview on Sept. 29, with broader release planned in the coming days. Axios described the appeal as AI judgment without unlimited freedom, suitable for narrow repetitive choices such as approving or rejecting a request.

OpenAI also detailed Bedrock Managed Agents, built with Amazon. The offering takes core Agents API capabilities and adapts them to run natively in Amazon Web Services and connect with AWS resources, so developers can build OpenAI agents that run entirely inside AWS, according to the DevDay 2026 Recap from OpenAI.

On trust, OpenAI introduced Private Intelligence for business use. Zero Data Retention with Private Safety Processing allows automated safety checks without giving OpenAI personnel access to underlying content, according to the same recap. A Private Inference preview combining confidential computing with verifiable controls was planned for fall.

Plugins turn ChatGPT into a launch surface

OpenAI said it is opening the internal platform used to build ChatGPT features so outside developers can ship native experiences to its user base, according to the DevDay 2026 Recap from OpenAI.

Plugin extensions can add a sidebar home, interactive panels that work alongside conversation and viewers for file types a product supports, according to that recap. The extensions are available to all plans. Figma and Adobe demonstrated integrations, according to the developer-community summary.

Creation and discovery tools were also updated. Plugin Creator assists building, while a redesigned submission flow offers clearer feedback, according to the DevDay 2026 Recap from OpenAI. Improved ranking and in-conversation recommendations are intended to aid discovery. Other distribution pieces cited in the community summary include Sign in with ChatGPT for other products, usage portability through 16 launch partners, an OpenAI Marketplace for eligible enterprise customers to apply commitments to partner products, and Sites that can use plugins with per-person data connections.

Shared workspaces, meetings and messaging

For teamwork, ChatGPT Space provides a persistent place where teams, ChatGPT and Dots share files and project context, according to Axios. Axios described the pitch as blending elements of Slack, Notion and Google Drive with embedded agents.

Pages are editable documents where teammates and agents collaborate, including on charts and interactive tools, while collaborative slides were described as coming in the next few weeks, according to the developer-community summary. Teams and Team Tasks support shared work and recurring assignments.

ChatGPT also meets users in existing tools. It can respond to @ChatGPT mentions in Slack and Microsoft Teams and use connected tools inside those conversations, according to that summary. A Meetings plugin generates notes and action items and was in beta on macOS for Pro and Business users. Shareable profiles gather a user's Sites and plugins on one page.

Safety, hardware and what OpenAI did not release

Safety loomed over the product news. CNBC reported that OpenAI has faced pressure after disclosing incidents of unintended behavior and holding back its latest Astra model. Earlier in September, the company endorsed a call to slow advanced-model work to address risks, according to CNBC.

Axios reported that OpenAI said it would not release GPT-6.1 Astra because of security concerns. Altman explained the withholding choice during the keynote, according to CNBC's agenda of remarks. Friar said OpenAI is speaking more about safety because of real-world events, adding that the lab will show model progress in some cases and restrain frontier releases in others, according to CNBC.

Altman also looked ahead. He teased future AI hardware as "something that's worth waiting for" and said OpenAI is working on something similar to Nvidia's agent-safety platform, according to CNBC. He called Meta's Muse a "nice product" while positioning Dots as a deeper personal agent, according to CNBC.

Chairman Bret Taylor was scheduled to discuss enterprise AI scaling later in the week at CNBC's AI Forum, according to CNBC, underscoring how DevDay's consumer and developer launches feed OpenAI's business push.

Sources referenced