Hi @d8rt8v - we've been working on a number of improvements around reasoning models and will figure out a solution to this. In the meantime, you can use include_original_response (or include_raw_response on the latest release) to fetch them (but you'll have to parse).

@GabrielBianconi for our discussion in slack

I already migrate to use t0 native API. But another issue I found is if I try to call a OpenAI compatible LLM provider, and they require thought content, then we even can't call them

I think it's more critical as include_raw_response still not work in our case.

One such model is kimi-2.5 https://platform.moonshot.ai/docs/api/chat#chat-completion

Hi @wangfenjin - thanks for the details. I'll take a look at Moonshot's reasoning API as I continue to expand the reasoning support on TensorZero.

In the meantime - though I agree it's not ergonomic - you can insert reasoning_content in follow-up inferences using extra_body. Let me know if you'd like an example. Thanks!

Understand, thanks!

@wangfenjin upgrading this to an issue, thank you for the feedback #5969