OpenAI DevDay 2026: Launch of Dots personal agent

OpenAI launched Dots, an always-on personal agent powered by GPT-6 Astra, at DevDay 2026 alongside the ChatGPT Space workspace for paid business users. It also introduced GPT-6.1 Sol at one-fifth of Astra's price and an OpenAI Marketplace with 30-plus partners.
Key insights
Dots: an always-on personal agent that acts first and asks later
At DevDay 2026 at Fort Mason in San Francisco, CEO Sam Altman introduced Dots as an always-on personal agent running on GPT-6 Astra, which he called OpenAI's best-aligned model. Each Dot has its own cloud computer and browser, writes code, runs tests, and inherits user permissions across more than 4,000 connected apps, with SMS and phone contact planned. Dots proactively push work forward and request approval only at critical nodes. Demonstrated tasks ranged from migrating an application off a deprecated API with traced dependencies and a submitted pull request, to revising board materials, replacing a canceled wedding cake vendor and resolving a scheduling conflict with a daughter's performance. Altman said his own Dot reviews overnight messages each morning and flags urgent items, giving back some of his attention.
ChatGPT Space adds a shared workspace where agents work on pages
ChatGPT Space organizes pages and files in one workspace where users write plans, do research and generate images and data, with Dots working directly on pages. Mentioning a Dot in a comment assigns it the task, and pages can carry standing instructions such as monitoring a Slack channel and posting updates. Product lead Holly demonstrated slash commands inserting interactive charts, tables and runnable prototypes, then had her Dot Dotty convert a dataset to a filterable bar chart refreshed hourly and fill an FAQ with onboarding data a engineer had mentioned in a Slack DM, resolved through Dotty's access to her context. Presentation formats designed for agent readability and collaborative editing with Dots are planned.
Inside OpenAI, Dots fix dozens of bugs daily and write parts of Dots
Holly reported that Dots hold their own identities in OpenAI's Slack, where employees route @-mentioned requests to their Dots and add them to group chats from the start. Engineers' Dots pick up user bugs posted with session IDs in feedback channels, investigate and submit fix pull requests, fixing dozens of bugs daily, and substantial parts of the Dots product itself are written by Dots. A simulated launch-eve demo for a playlist app showed Dotty rescheduling a moved review meeting, reading overnight tester feedback, building a redesign with Codex on a laptop and iPhone simulator, and submitting a PR, with a live voice demo stalling while Dotty kept replying that it was still checking.
Specialist Dots bring company-run virtual coworkers with Microsoft integration
Dots and Space shipped the same day to ChatGPT Pro, Business Premium and Enterprise users, with Dot conversations excluded from usage quotas. Enterprise customers can preview Specialist Dots, virtual coworkers configured company-wide for high-volume functions such as accounting, marketing and legal, which receive goals and background from the company and produce outputs that are reviewed with feedback shared across the organization. OpenAI is also partnering with Microsoft to connect Specialist Dots into Agent 365, Microsoft's enterprise agent management tool, letting companies administer them through existing Microsoft tooling.
GPT-6.1 Sol, UltraFast and a $500 Pro 500 plan address cost and speed
Responding to user demands for cheaper and faster models, OpenAI released GPT-6.1 Sol with capabilities close to Astra at one-fifth the price, cached input 95% cheaper than standard input, and, per Altman, intelligence exceeding Astra in some respects as the developer daily workhorse. The new UltraFast tier delivers 8x speed at 6x price, 300 tokens per second, across API, ChatGPT and Codex, currently paired with Astra and later Sol; a live race showed UltraFast finishing a rocket-building prompt before standard speed. The $500 Pro 500 subscription offers 25x Plus quota, UltraFast access and Sign in with ChatGPT in partner apps, while Pro 200 reopened. A Decisions API preview restricts the Luna model to predefined choices, compressing responses under one second for routing, classification and robot control.
Codex goes fully cloud with an open harness, security scanning and Agents API
OpenAI open-sourced the Codex harness that simultaneously powers Codex, ChatGPT Work and Dots, and made Codex fully cloud-based so tasks move between phone, browser and desktop without interruption. Codex Security Cloud continuously searches cloud environments for vulnerabilities and prepares verified fixes for human review, adding automatic deduplication, scheduled scans and a new interface. The Agents API entered public beta with the harness, hosting, memory, multi-agent control and computer use, demonstrated by a website-testing agent that opens a browser and tests flows itself. OpenAI and AWS launched OpenAI-powered hosted agents on Bedrock, and a Private Intelligence preview added zero data retention and private inference for sensitive workloads. The Responses API grew 100x in a year at 99%+ reliability, with first-token latency down 45% and tool calls over 30% faster.
ChatGPT becomes a distribution platform with sign-in, plugins and Marketplace
With roughly 1.2 billion weekly ChatGPT users, OpenAI launched Sign in with ChatGPT, letting users spend tokens included in their own plans inside 16 launch-partner apps so developers no longer front model costs for new users. Plugin extensions embed editors, dashboards and entire workspaces natively in ChatGPT and Codex, demonstrated by a meetings app that turns calendar entries into shared Spaces, Figma design review, and Photoshop features from Adobe; ChatGPT sites can now attach plugins and per-user data. Discovery expands through in-conversation plugin suggestions, and review is simplified with progress tracking and human re-review. The OpenAI Marketplace opened with more than 30 partners including CodeRabbit, Notion and Vercel, purchasable with committed OpenAI spend and including open models via Baseten.
Altman closes with a Renaissance framing and another quota reset
Closing DevDay, Sam Altman said he dislikes comparing AI to a new industrial revolution because it casts people as cogs in a giant machine spinning ever faster, adding that some parts of life cannot and should not be automated. If done right, he argued, AI would resemble a new Renaissance, giving people more control over their lives and more tools to create, learn and explore. The day also included sessions on OpenAI's open-source commitments and a Codex game studio where attendees built retro games on a limited-edition chromatic computer, and Altman and Thibault pressed a button resetting usage quotas worldwide, a practice Altman joked had nearly earned the company the name Reset Company.