OpenAIAnthropicGooglexAIPerplexity
← Back to all updates
Published ·Anthropic• Minor

Messages API now reports usage.output_tokens_details.thinking_tokens for billed output tokens used as extended thinking.

The read

Billing audits can isolate extended thinking because the response separates thinking tokens inside billed output tokens.

The frame

For streaming builds, the token breakdown arrives only on the final message_delta event and needs no beta header.

Primary source
Anthropic
Keep reading
See what Nova3 builds
Related updates

This is the world we build in. It moves this fast. We move with it.

Bring us your project