Using the /responses API with an OpenAI Compatible Provider

[mstykow]

Hi everyone!

I'm using an OpenAI Compatible Provider to communicate via Vercel AI SDK with a self-hosted gpt-oss-120b.

It seems that the SDK sends all requests via the /completions API rather than the newer /responses API which is recommended by OpenAI for all of its newer models (see also here), including gpt-oss.

Is there some way to tell the OpenAI Compatible Provider to use the newer API? I couldn't find any discussions or issues around this topic but also find it hard to imagine that I'm the only one wondering about this.

Thank you for any pointers!

[foxxgreeley]

Looking for an update on this as well. I know that passing in gpt-5 models will use responses but I don't want to use this model due to reasoning time.

[github-actions[bot]]

This discussion was automatically locked because it has not been updated in over 30 days. If you still have questions about this topic, please ask us at community.vercel.com/ai-sdk