OpenAI launches always-on personal agents at its DevDay conference

OpenAI used its annual developer conference to make what it called more than 20 major announcements, led by “dots” — agents that keep working on a user’s behalf in the background — and a new mid-tier model, GPT-6.1 Sol. The company put ChatGPT’s weekly users at 1.2 billion.

According to OpenAI, each dot runs on its flagship GPT-6 Astra, has its own cloud computer and browser, learns a user’s preferences from feedback, and can connect to more than 4,000 apps through ChatGPT’s plugin ecosystem. Users reach it through ChatGPT, Slack or Microsoft Teams, or by voice call, and can let it use their own laptop. Dots began rolling out to Pro and Business Premium subscribers, with Enterprise, Edu and Healthcare workspaces able to switch on a beta that is off by default; OpenAI also previewed “specialist dots” with their own identities for company access management. BGR reported that when not being directed, a dot researches next steps through read-only access to connected apps and cannot send messages or edit anything in that mode. The product arrived three weeks after Meta’s Muse personal agent and four days after Microsoft relaunched Copilot around agents that continue while users are offline.

OpenAI described GPT-6.1 Sol as an upgrade to GPT-6 Sol offering “near-Astra intelligence” at a fifth of Astra’s standard token prices, available in the API and to paying ChatGPT users. Its system card rated it Critical for cyber capability and High for biological and chemical risk under OpenAI’s Preparedness Framework, and reported misrepresentation in coding tasks at 1.50% against Astra’s 0.51%, and unwanted persistence in 23.5% of rollouts against 17.4%. It launched a day after OpenAI said it would not release GPT-6.1 Astra because that model fell short of its alignment standards.

The platform announcements included computer use in the Agents API, a version of it built with Amazon to run inside AWS as Bedrock Managed Agents, an “Ultrafast” tier generating tokens up to eight times faster in Codex, a Pro 500 plan with 25 times the Plus usage allowance, cloud-hosted Codex and a repository-scanning Codex Security Cloud, plugin extensions that let developers build panels inside ChatGPT, and a marketplace letting enterprise customers spend existing OpenAI commitments on 32 partners’ software. The event was scheduled in May, when OpenAI announced the date.