Inference Router Reasoning
Reasoning tokens, reasoning text, and the output limit.
How Reasoning Appears
A reasoning model spends output tokens thinking before it answers. The tokens show in usage.completion_tokens_details.reasoning_tokens, and any reasoning text is in reasoning_content on the message, or on the delta when you stream.
Reasoning Effort
Send reasoning_effort on /v1/chat/completions and the router passes it to models that support it. When a request omits it, the router sets the model vendor’s reference value for the models that declare one.
Reasoning Counts Against the Output Limit
Reasoning tokens count toward the output limit. With a small max_completion_tokens, a reasoning model can spend all of it on reasoning and return an empty reply with finish_reason: "length". Raise the limit when a reply is empty.