OpenAI is upgrading Operator, its AI agent, with the o3 model to enhance reasoning and safety. Operator is designed to autonomously browse the web and interact with software within a cloud-hosted virtual machine to fulfill user requests. The update replaces the existing GPT-4o-based model with one based on o3, the latest in OpenAI’s series of “reasoning” models. According to OpenAI, by several benchmarks, o3 is particularly effective at tasks requiring mathematical ability and reasoning.

OpenAI mentioned in a blog post that they are replacing the existing GPT‑4o-based model for Operator with a version based on OpenAI o3, but the API version of Operator will remain based on 4o. Operator is one of many agentic tools developed by AI firms competing to build agents capable of reliably performing digital tasks with minimal supervision.

Other companies are also developing similar agents. Google offers an agent through its Gemini API that can browse the web and act on users’ behalf, and it also offers a consumer-facing version called Mariner. Anthropic’s models can also perform various computer tasks, including opening files and navigating webpages.

The upgraded Operator model, known as o3 Operator, has been “fine-tuned with additional safety data for computer use.” Datasets were used to “teach the model (OpenAI’s) decision boundaries on confirmations and refusals.” OpenAI has released a technical report detailing o3 Operator’s performance in safety evaluations. The new model is less likely to carry out illicit activities, search for sensitive personal data, or fall prey to prompt injection compared to the GPT-4o version. o3 Operator employs the same multi-layered approach to safety used for the 4o version. While o3 Operator inherits o3’s coding capabilities, it does not have native access to a coding environment or terminal.