OpenAI is updating the AI model powering Operator, an autonomous web browsing agent that can fulfill user requests. The new model, called o3 Operator, is based on one of OpenAI’s latest “reasoning” models and offers significant improvements over its predecessor, GPT-4o. The o3-based model excels in math and reasoning tasks, according to benchmarks.
The upgrade comes as AI companies continue to develop sophisticated agents that can perform tasks without human supervision. Other notable agentic tools include Google’s Gemini API and Anthropic’s models, which can browse the web, open files, and navigate web pages.
OpenAI has fine-tuned the new model with additional safety data, including datasets designed to teach the model about decision boundaries on confirmations and refusals. The company has released a technical report showing o3 Operator’s performance on specific safety evaluations, which indicate significant improvements over its predecessor.
The updated model is less likely to refuse tasks that involve illicit activities or search for sensitive personal data, and is also more resistant to prompt injection attacks. While the new model inherits some coding capabilities from the o3 series, it does not have native access to a coding environment or terminal.
Source: https://techcrunch.com/2025/05/23/openai-upgrades-the-ai-model-powering-its-operator-agent