Completion Tokens and Thinking Tokens in OpenAIv2 Instrumentation

What problem do you want to solve?

The semconv and spans generated by the openai instrumentation include input tokens and output tokens. These are a great start, but I would like to propose adding completion tokens to these as well.

Completion tokens are a very important piece of information for debugging and evaluating agentic systems. Efficient prompt usage is an important element when building agent systems and optimizing cost.

Describe the solution you'd like

I would like to add a new attribute gen_ai.usage.cached_tokens to the span generated by the openai instrumentation. The cached tokens can be populated from the Completion object returned by the API.

My guess is this would need to be added into the gen-ai semantic conventions. I think a case could be made that completion tokens could fit into the semantic conventions since other providers also have some concept of completions tokens.

Describe alternatives you've considered

No response

Additional Context

No response

Would you like to implement a fix?

None

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.}

Start by locating the OpenAIv2 instrumentation and reviewing the linked GenAI semantic conventions and OpenAI Completion object documentation. Clarify whether cached, completion, and thinking tokens are in scope, then determine how the selected attribute should be represented in generated spans and what instrumentation behavior would verify completion.