OpenAI upgrades Operator to the powerful o3 model for smarter AI tasks
OpenAI is set to give its AI agent, Operator, a significant upgrade by switching to a new model that has only recently joined the company’s series of “reasoning” models.
The company is moving from a custom version based on GPT‑4o to o3, which only emerged earlier this year and has proven to be a real whizz at math and reasoning tasks.
In a blog post today, the company said: “We are replacing the existing GPT‑4o-based model for Operator with a version based on OpenAI o3. The API version of Operator will continue to use the GPT‑4o model.”
The move comes as the race heats up between the major AI companies to create sophisticated agents that can take care of tasks on behalf of users with little or no input. Google is pushing hard with its Gemini API, while Anthropic is also making great strides.
The new o3 Operator has been fine-tuned with extra safety data to help it navigate the boundaries of what it should and shouldn’t do when making decisions.
A technical report accompanying the announcement says the new model is less likely to refuse to take part in illicit activities and less likely to fall victim to prompt injection attacks that seek to compromise the agent.
OpenAI said that while o3 Operator remains a whizz at coding, it doesn’t have access to a coding environment or a terminal, meaning the multi-layered safety approach of the previous model remains in place.
