OpenAI took the stage at DevDay 2026 in San Francisco on Tuesday and did exactly what it had teased: more than 20 product announcements in a single keynote. The headline is Dots — always-on AI agents powered by GPT-6 Astra that keep working on your behalf around the clock, each with its own cloud computer and browser, connected to more than 4,000 apps and plugins. "Today, we introduced agents that can take on ongoing responsibilities and new ways for people and AI to work together," OpenAI wrote in its DevDay summary, describing ChatGPT as a shared surface for its 1.2 billion weekly users.

Dots are OpenAI's direct answer to the personal-agent land grab. Announced live in the keynote, each Dot learns from your feedback, proactively surfaces useful work from your connected apps, and can be reached in ChatGPT, Slack, and Teams. They roll out now to Pro, Business Premium, and Enterprise plans in eligible markets, with more users onboarding soon — and, notably, without drawing down your regular usage.

#ChatGPT Space: a shared home for people and agents

The connective tissue is ChatGPT Space, a collaborative workspace where people, ChatGPT, and Dots work on the same material. As VentureBeat reports, Space becomes the shared home for Pages and files — interactive documents combining text, images, charts, and dashboards that employees and AI edit together in real time, replacing the old Library for eligible users. Connected tools like Google Drive and Slack supply context; collaborative slides and spreadsheets are coming.

Recurring work gets automations: delegations that gather information and act on schedules or external events, so employees and multiple agents hand work back and forth against the same project context instead of operating as isolated one-to-one assistants.

Illustration of glowing AI agent spheres working at a computer screen, connected by threads of light to app icons
AI-generated illustration for AI Frontier Post.

#GPT-6.1 Sol and Ultrafast: cheaper and faster

On the model side, OpenAI launched GPT-6.1 Sol, which it describes as offering near-Astra intelligence on agentic coding, computer use, and professional work at a fifth of Astra's standard input and output token prices. It's available now across the API and all Plus, Pro, Business, Enterprise, and Edu tiers. OpenAI's own benchmark figures show Sol scoring 75.2% on DeepSWE v1.1 (vs 68.8% for GPT-6 Sol) at roughly 76% lower cost per task, and 31.7% on AutomationBench — while reaching 71.4% on OSWorld 2.0's offline set at about one-seventh of Astra's cost per task.

For speed, Ultrafast delivers up to 8× faster token generation in Codex — reaching 300 tokens per second — and up to 6× in the API.

#The developer stack: Agents API, Marketplace, Codex

Developers get the Agents API in public beta: a managed agent runtime with memory, multi-agent capabilities, environments, sessions, tools, and MCP support, plus computer use that runs in an OpenAI-hosted browser. OpenAI also expanded Codex — cloud environments roll out to Plus, Pro, Business, and Enterprise so teams share setups, and the Codex CLI gains parallel-agent controls.

And there's a Marketplace with more than 30 launch partners, where developers can distribute plugin extensions that run natively inside ChatGPT — including Adobe Photoshop features and Figma design review — with "Sign in with ChatGPT" letting users bring their subscription usage into third-party products.

Illustration of a shared digital workspace with floating living documents and glowing agent collaborators
AI-generated illustration for AI Frontier Post.

#What to watch

Twenty announcements in one keynote is a lot — but the shape of the bet is clear. OpenAI no longer wants to be the box you type into; it wants to be the team you never hire. Whether users trust an always-on agent with a browser, a calendar, and a Slack workspace is the question Dots will answer in the wild.