Area(s)

area:gen-ai

Propose new conventions

Since reasoning tokens are a internal part of a chat run, they are neither input nor really output, it would help to clarify what the model did when the spec also adds reasoning tokens, as a separate category, next to input and output.

Alternatives could be thinking, which is what Google uses, for Gemini models, but probably reasoning makes more sense, since there are also reasoning message as input/output content conventions.

So concretely:
add:
gen_ai.usage.reasoning_tokens
as a optional span attributes for chat

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}