OpenAI has upgraded the AI model powering its Operator agent from a custom GPT-4o-based model to a new version based on OpenAI's o3 model. This update enhances the agent's persistence, accuracy, reasoning capabilities, and safety features, particularly in math and autonomous web browsing tasks. The o3 Operator model was fine-tuned with additional safety data for computer use, improving task success rates and delivering clearer, more thorough, and better-structured responses. The Operator remains available exclusively to ChatGPT Pro subscribers, who pay $200 monthly. The upgrade is viewed as a substantial improvement over the previous GPT-4o model, with expectations of further enhancements such as longer timeouts and scheduling. Additionally, there are indications that Operator may soon gain more code-related capabilities, including support for Jupyter and terminal servers in specific builds.
I’ve noticed Operator might get more code-related capabilities soon - the container shows enabled Jupyter and terminal servers in specific builds like “caas-appberry” and “operator-debian-alpha”, plus the web app already has a “code” content type https://t.co/EReMulhivF
DuckDuckGo mejora su IA con GPT-4o mini sin comprometer tu privacidad https://t.co/UqrqYUEo1L
You can run full browser automations for AI agents without worrying about Chrome, Puppeteer, or infrastructure. Steel is an open-source browser API that wraps Chrome, manages sessions, handles proxies, and exposes everything via a REST API or SDKs. https://t.co/9XKUl7zPxg