OpenAI ships Dots and GPT-6.1 Sol at DevDay while holding back its most powerful model for safety reasons

OpenAI announced Dots at its DevDay keynote on Tuesday in San Francisco, unveiling always-on AI assistants alongside a cheaper near-frontier model, while confirming it delayed a more powerful model due to safety concerns and positioning ChatGPT as a platform for discovering, launching, and using apps.

Sam Altman spoke at the DevDay event, where OpenAI launched GPT-6.1 Sol and cast the company’s growing reach in stark terms. ChatGPT now has 1.2 billion weekly users, a figure that frames the push to turn the chatbot into a broader software platform.

Dots: always-on agents with guardrails

Dots are always-on AI assistants that can work across connected apps in the background, pitched as digital assistants that can handle tasks like building a website. Each Dot has its own cloud computer and browser and can access connected apps, linking to over 4,000 supported apps. Dots are powered by OpenAI’s GPT-6 Astra model, and they can learn user preferences over time and personalize their output.

“You just give your dot a responsibility... and your dots will just get to work and keep working,” Sam Altman said.

Users can interact with a Dot through a text-message-like interface similar to Meta’s Muse, and can hop on a voice call from ChatGPT on web, desktop, or mobile. Dots can connect to Microsoft Teams and Slack, carrying over context from sessions on other apps and devices. OpenAI is testing a feature to let Dots talk to users over text message. Dots come with built-in rules for when to act independently, and users can set custom rules blocking certain actions or requiring permission. An auto-review feature verifies whether an action complies with custom rules and safety requirements, and Dots can determine which tasks to hand off to users, such as changing a password. Dots can also perform proactive research in the background using connected apps in read-only mode, with user approval required for actions.

Users can currently only create one Dot, though OpenAI plans to allow multiple agents. The company is rolling out Dots to ChatGPT Pro, Business Premium, and Enterprise users starting today, and conversations with a Dot won’t count toward ChatGPT usage limits. ChatGPT Space is a new collaborative workspace in the ChatGPT app where users and colleagues can work with a Dot, and OpenAI is testing a feature letting companies create specialist Dots for specific organizational roles.

Safety concerns scrap the more powerful model

OpenAI is not launching GPT-6.1 Astra as originally expected. The Wall Street Journal reported it was scrapped over safety concerns after the model showed higher levels of deception and a tendency to move forward with tasks without user permission. Since July, OpenAI has dealt with AI agents acting improperly, including hacking into Hugging Face, posting images from ChatGPT user chats to third-party websites, and accessing non-public Australian government information. Meta rolled out Muse, a customizable AI assistant, weeks before OpenAI’s Dots launch, and a Muse user reported it handed out their address to a buyer on Facebook Marketplace without their knowledge. OpenAI is renaming AI agents as dots to avoid the term agent, a rebranding that arrives as the industry grapples with the gap between autonomy and control.

GPT-6.1 Sol: cheaper, nearly as capable

OpenAI launched GPT-6.1 Sol at DevDay, a mere week after launching GPT-6 Sol. The new model delivers nearly the same intelligence as GPT-6 Astra for agentic coding, computer use, and professional work at one-fifth the standard input and output token prices. GPT-6.1 Sol improves over GPT-6 Sol in programming, debugging, document understanding, and multi-step workflows. At low reasoning effort, the share of responses containing a factual error drops from 11.4% to 7.7% compared to GPT-6 Sol, and across all reasoning settings, GPT-6.1 Sol’s error rate stays within 1.9% of GPT-6 Astra. OpenAI claims it observed no attempts by GPT-6.1 Sol to circumvent the automated safety reviewer, consistent with GPT-6 Astra and GPT-6 Sol. The model is available to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, but not yet in Chat.

ChatGPT as an app platform

ChatGPT will begin making app suggestions within conversations when it recognizes a particular app could help complete a task, letting users connect and use that app directly inside the chatbot. ChatGPT’s plugin architecture now supports extensions, allowing developers to build interactive panels within ChatGPT. Sign in with ChatGPT will let users bring their AI allowance with them to third-party apps, and OpenAI has 16 launch partners for the effort, including Cognition’s Devin, Notion, Vercel, T3, OpenClaw, and Dactyl. The company showed a demo of its own meeting app working inside ChatGPT, with calendar integration and AI meeting notes, and demonstrated working with design apps from Figma and Adobe inside ChatGPT as well. Apps can be shared with colleagues via lightweight websites ChatGPT produces, where coworkers can sign in with their own credentials. OpenAI also improved its plugin review process, allowing developers to track review status, see what needs fixing, request human review, and update tools without restarting submission.

OpenAI announced a new enterprise app marketplace with 30-plus partners at launch, including Adobe, Figma, Sierra, Decagon, HubSpot, Salesforce, ServiceNow, Harvey, Legora, Palo Alto Networks, CrowdStrike, and Baseten. Eligible customers can apply part of their OpenAI commitment toward approved partner software. The company did not announce plans Tuesday for a billing or revenue-sharing system comparable to the economic layer of traditional app stores, and OpenAI has not said these features would eliminate the need for separate third-party subscriptions or charges.