Understanding the Large Difference in Token Count Between Session and Trace Counts

·Mar 06, 2026 09:06 AM·

Hi ! I have a question regarding the "Token Count" indicator for session and traces. I have a trace that indicate 700k tokens used with 131k tokens in input and 26k tokens in output (completion token in Arize language if I got that right). Why is there a this big difference ?

3 comments

· Sorted by Oldest

Tristan L.
·
·
🔒[private user] does that means that it could be the use of tools ?
Rick M.
·
·
Tristan L. another potential source to the disparity could be prompt caching. For example, I often see much higher token count in Claude Code than's actually going over the wire. https://platform.claude.com/docs/en/build-with-claude/prompt-caching https://developers.openai.com/api/docs/guides/prompt-caching/
Tristan L.
·
·
Noted, thank you for the precision !

Tristan L.
·
·
🔒[private user] does that means that it could be the use of tools ?
Rick M.
·
·
Tristan L. another potential source to the disparity could be prompt caching. For example, I often see much higher token count in Claude Code than's actually going over the wire. https://platform.claude.com/docs/en/build-with-claude/prompt-caching https://developers.openai.com/api/docs/guides/prompt-caching/
Tristan L.
·
·
Noted, thank you for the precision !