https://docs.ollama.com/api/usage (2026-09-11).
Cached prompt tokens are a subset of prompt_eval_count. Older servers omit
the optional cache field: absence is unknown, not a measured miss or zero.
Nanosecond timings are not token counts, prices or cache-write evidence.
https://docs.ollama.com/api/usage (2026-09-11). Cached prompt tokens are a subset of prompt_eval_count. Older servers omit the optional cache field: absence is unknown, not a measured miss or zero. Nanosecond timings are not token counts, prices or cache-write evidence.