release

Addendum to OpenAI o3 and o4-mini system card: OpenAI o3 Operator

For developers building AI workflows, this change means the Operator product will now run on a different model, potentially altering automation outcomes, while the API remains stable—requiring separate testing for each interface.

OpenAI Blog·May 22, 2025·1 min readrelease

releaseAddendum to OpenAI o3 and o4-mini system card: OpenAI o3 Operator

openai.com

What happened

OpenAI has updated the model powering its Operator tool in a recent addendum to the o3 and o4-mini system card. The company announced that Operator now uses a version based on OpenAI o3, replacing the previous GPT-4o-based model. However, the API version of Operator remains on GPT-4o, meaning developers interacting through the API will not see any immediate change. This update comes as part of OpenAI's broader rollout of the o3 reasoning model, which was introduced earlier this year. For builders embedding Operator into their workflows via the API, the transition is currently transparent, but product users may notice different behavior due to o3's altered reasoning patterns. The change highlights OpenAI's strategy of iterating on its frontier models while maintaining API stability. Developers should monitor for any shift in output quality or reliability when using the Operator product, and consider testing their existing automations against the new model to ensure consistency.

Key takeaways

OpenAI replaced the GPT-4o model in the Operator product with a version based on o3.
The API version of Operator continues to use GPT-4o, with no change announced.
This update is part of the official o3 system card addendum.
The switch may affect Operator's performance and reasoning capabilities in the product.
Developers using the API should note no immediate impact, but may want to test product behavior.