Question

A developer is building a chat feature that streams Claude's responses to the user in real time. After the stream completes, they need to log the exact number of input and output tokens consumed so they can track per-conversation cost. Where in the streaming flow should they look to get this final token usage?

Accepted Answer

The final message_delta / message_stop events at the end of the stream, which report the cumulative usage

Answer

The very first message_start event, which contains the complete final usage totals before any text is generated

Answer

Each individual content_block_delta event, summing the token counts they each carry

Answer

Streaming responses do not include usage data, so a separate non-streaming request must be repeated to get token counts