OpenAI used its DevDay 2026 conference in San Francisco on Tuesday to pitch the idea that your next coworker could be an AI with its own computer, announcing more than 20 launches. They range from always-on agents called "Dots" to a cheaper workhorse model, confirmation of the rumored $500-a-month plan, and a shared workspace where humans and bots edit the same documents.

The announcements reinforced the company's framing of AI as shifting from a chatbot you consult to a worker you assign things to—one with its own computer, connections to your work apps, and the ability to keep going while you sleep. The rollout also drew sharp lines by plan, with Dots, Space, and the upgraded Agents API all skipping Plus, the $20-a-month tier.

The event came a day after OpenAI pulled its next flagship, GPT-6.1 Astra, because it did not meet safety standards. That followed a summer in which OpenAI's agents breached the open-source repository Hugging Face, the Australian and U.S. governments, and private companies.

Dots: AI agents that never clock out

Dots are the headliner. OpenAI calls them "remarkably capable, always-on agents", meaning AI that takes actions on its own instead of just answering questions, and keeps working after you close your laptop. It is the same premise the open-source AI agent platforms Hermes and Openclaw have been developing for months.

Each dot runs on GPT-6 Astra, OpenAI's flagship model, and gets its own cloud computer and web browser. It can plug into more than 4,000 apps and learns your preferences from feedback. You can reach it in ChatGPT, on a voice call, or in Slack and Teams, with texting coming soon.

Dots follow built-in rules on when to act alone, and jobs like changing a password always stay with the human. The first dot is included with Pro and Business Premium in eligible markets, with Free and Plus users left out.

ChatGPT Space: where humans and dots share files

ChatGPT Space is a shared workspace where teammates, ChatGPT, and each person's dot work from the same pile of project knowledge—like a shared drive plus group chat, except some of the members are software. It introduces Pages, a new kind of document built for people and agents to edit together.

In the keynote demo, OpenAI's Holly Li tagged coworkers and her dot inside Space and handed out chores: write an FAQ, update her, and comb through her Slack DMs. Collaborative slides that export to PowerPoint or Google Slides are coming in the next few weeks.

Space is available on Pro, Business, and Enterprise plans, on desktop and web. Plus, the $20-a-month plan, is not on the list. Mobile can read and share Pages, but creating them there is coming soon.

GPT-6.1 Sol: near-flagship performance at a fifth of the price

GPT-6.1 Sol is OpenAI's new workhorse model, arriving a week after GPT-6 Sol. OpenAI says it nearly matches GPT-6 Astra on agentic coding, computer use, and professional work at one-fifth of Astra's token prices. Tokens are the chunks of text a model reads and writes, and they are how AI companies bill.

It lands in a bruising week: OpenAI canceled the October launch of GPT-6.1 Astra, and CEO Sam Altman told CNBC he would address it later.

Sol is live in the API and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users.

Ultrafast: paying more to go faster

Ultrafast is OpenAI's premium speed tier. It generates tokens up to 8x faster in Codex—about 300 tokens per second—and up to 6x in the API, with users paying up to 6x the price of standard usage. Speed matters because agents take dozens of small steps, and every step waits on the model.

Astra Ultrafast is live now, and a Sol version is coming soon. It is available on the $500-a-month Pro 500 and Enterprise plans.

The live Ultrafast demo failed several times on stage. The tier has run on Cerebras chips, built as single dinner-plate-sized processors, since OpenAI's August preview.

Decisions API: quick answers to multiple-choice questions

The Decisions API points Luna, OpenAI's cheapest model, at questions with a fixed list of possible answers. Developers send text or images and get back a pick they can use to sort content, route a request, or choose an agent's next move—think of a bouncer, not a philosopher. The model is not necessarily thinking more or better; OpenAI pitches it as near-instant decision-making to improve reliability on the API. It is in limited preview, with a broad release planned in the coming days.

Codex harness: the engine under dots goes open source

A harness is the software wrapper that lets an AI model use tools, run commands, and keep track of a task. At DevDay, OpenAI said the Codex harness, built for its coding agent, now powers dots and is open source, meaning anyone can read and reuse the code. OpenAI's docs list the Codex CLI, its SDK, and its app server as open-source projects on GitHub.

Open code lets outsiders inspect how an agent is wired, which matters after months of agents breaking out of their test environments.

Codex Security Cloud: an AI that hunts bugs while you sleep

Codex Security Cloud points OpenAI's coding agent at defense. It can scan entire GitHub repositories on demand or on a schedule, check every new commit, investigate findings, remove duplicates, and prepare fixes in the cloud, even with your laptop closed.

It includes models offered through Daybreak Blue, part of the security push OpenAI launched in May, with no separate application. It is available to Pro, Business, Enterprise, and Edu users.

OpenAI has classed Astra as its first model with critical cyber capability, able to find and chain zero-days—flaws nobody has patched yet.

Agents API: computer use for developers

The Agents API now supports computer use, meaning an AI that clicks, types, and navigates software the way a person would. It also brings Codex's multi-agent tools, tool search, and context compaction—a way to shrink a long task history so the model doesn't lose the plot—into developers' own products. OpenAI runs the underlying infrastructure, so developers don't host the machinery themselves.

It is available through the API and in Codex and ChatGPT Work on Pro 500 and Enterprise. OpenAI also announced Bedrock Managed Agents with Amazon, to run OpenAI agents entirely inside AWS. Plus and free users won't have this feature.

Private Intelligence: AI for companies that can't share secrets

Private Intelligence is OpenAI's pitch to companies nervous about handing over sensitive data. It pairs zero data retention, a setup where the provider doesn't keep your prompts and outputs, with automated safety checks that run without OpenAI staff seeing the content.

Next comes Private Inference, a preview arriving this fall. Inference is the moment a model generates an answer, and OpenAI says it will run it on confidential computing—hardware that seals data inside a locked enclave even the cloud operator can't open—with verifiable controls.

For hospitals, banks, and law firms, that decides whether AI can touch client data at all. The approach differs from what other companies offer: Anthropic, for example, makes it impossible to opt out completely of sharing data and information. Users who agree give Anthropic permission to use their prompts for five years; those who don't, for 30 days.

Sign in with ChatGPT: bring your own subscription

Normally, an app that uses OpenAI models pays for tokens itself and folds the cost into its own price. Sign in with ChatGPT lets users spend their ChatGPT plan allowance inside other tools instead, and control how much each one can use.

OpenAI lists 16 partners, including Cognition's Devin, Notion, Vercel, T3, OpenClaw, and Dactyl. Plan usage works for Plus and Pro users in participating tools, and the sign-in itself is available globally. Eligible OpenAI usage counts toward your plan limits.

Plugin extensions: apps that live inside ChatGPT

Plugin extensions let outside developers build their own experiences inside ChatGPT, with a home in the sidebar, interactive panels next to the conversation, and viewers for the file types their product supports. OpenAI says it is opening the platform it uses to build its own features, on all plans.

The idea has history. OpenAI launched custom GPTs at its first DevDay in 2023 and turned ChatGPT into an app platform in October 2025. Those apps are now called plugins, and a redesigned directory ranks and recommends them, including inside conversations. Users choose which plugins to use and approve the access each one gets. With 1.2 billion weekly users, according to OpenAI, that is a very large storefront to build for.

OpenAI Marketplace: spend your AI budget on other software

OpenAI Marketplace lets eligible enterprise customers apply part of their existing OpenAI commitment—the prepaid spend they already agreed to—toward approved partner software.

The first 32 partners include Adobe and Figma for creative work, Salesforce, HubSpot, and ServiceNow for customer experience, Harvey and Legora for legal, and Palo Alto Networks and CrowdStrike for security. Enterprise customers can express interest now.

Elsewhere on the calendar, the Decisions API goes broad in the coming days, GPT-6.1 Sol Ultrafast follows soon after, and the Private Inference preview arrives this fall.