What would you like to see?

The Responses API for OpenAI is useful since it paves over the nuances of model differences without us having to create cases for each model - which changes so often.

The Agent provider for OpenAI is still using completions - breaking GPT-5, O-* outputs and generally causing incompatibilities with tools. Migrating to responses should resolve all of this.

This can have a ton of side-effects and should have no impact on other providers who override the baseURL or the OpenAI SDK we use where we leverage completions instead.