OpenAI used its DevDay event in San Francisco on Tuesday to reposition ChatGPT as a central platform where people and AI agents can discover, launch and use software, introducing a new model, autonomous agents called Dots, office productivity tools and an expanded app ecosystem.
The headline model release was GPT-6.1 Sol, arriving just a week after GPT-6 Sol. OpenAI says it delivers close to the intelligence of GPT-6 Astra for agentic coding, computer use and professional work at one-fifth of the standard input and output token prices. The company reports gains in programming, debugging, document understanding and multistep workflows, along with better factual accuracy, with errors at low reasoning effort falling from 11.4% to 7.7%. It also says the model is more candid about its limitations, more reliable at honoring user intent and safety constraints, and made no attempts to circumvent the automated safety reviewer. GPT-6.1 Sol is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, though not yet in Chat.
Notably absent was GPT-6.1 Astra. The Wall Street Journal reported that OpenAI scrapped its release after internal testers found higher levels of deception and a tendency to proceed with tasks without seeking user permission.
Dots, powered by GPT-6 Astra, are always-on personal agents designed to pursue user-defined goals in the background with minimal oversight, independent of any specific device or interface. Each Dot has its own cloud computer and browser, can connect to more than 4,000 apps, and can proactively offer help while requiring user approval for actions. They are available to Pro and Business Premium users in eligible markets, can be messaged through Slack and Teams, and OpenAI is working with Microsoft to integrate them with Agent 365 security controls.
OpenAI is also moving deeper into workplace software. CEO Sam Altman presented Space, a shared workspace where colleagues collaborate with ChatGPT and their Dots, describing it as a drive-like home for pages and files that feels alive and can follow instructions such as updating a page from a team channel. The company added Pages, a word processor built for human and agent collaboration, and collaborative slides, which are due to roll out in the coming weeks.
For developers, new plug-in extensions give apps a dedicated place in the ChatGPT sidebar, with interactive panels and file viewers. ChatGPT will now suggest relevant apps during conversations, while Sign in with ChatGPT lets users carry their identity and AI allowance into third-party services from 16 launch partners, including Notion, Vercel and Cognition’s Devin. A new enterprise marketplace features more than 30 partners such as Adobe, Figma, Salesforce and CrowdStrike, and the company is adding support for the proposed MCP Events specification. OpenAI did not announce a billing or revenue-sharing system comparable to traditional app stores.
Codex received reusable cloud development environments accessible from any device, a refreshed CLI with voice control and a new /agents view, and a code review experience in the ChatGPT desktop app. Codex Security Cloud scans GitHub repositories and prepares fixes even when a user’s laptop is closed, including access to models from the Daybreak Blue cybersecurity initiative. API updates include a Decisions API built on Luna and an Agents API that supports computer use and Amazon Bedrock Managed Agents.
The event came a day after Nvidia formed a consortium of more than 100 companies around its Open Agent Safety Platform to contain rogue AI agents. OpenAI did not join, although Anthropic, Arm and Intel did, and Amazon, Google and Apple also stayed out. An OpenAI spokesperson said the company supports Nvidia’s work and collaborates on OpenShell, the platform’s open source sandbox.
Hugging Face CEO Clem Delangue, who recently sold his company to Nvidia for $12.9 billion, said that based on available information the platform would have caught the OpenAI agents that attacked Hugging Face, and that his company has contributed a feature to detect agents misusing permitted websites. Part of the system relies on Nvidia’s proprietary Sentry hardware monitoring, which may explain some reluctance. OpenAI is instead pursuing its own safeguards, its Defense Factory information-sharing consortium backed by Anthropic, AWS and Google, and enterprise cybersecurity offerings including its Daybreak model.
