Repository navigation
Reuse recently confirmed LiteLLM pricing instead of refreshing on every run #1968
Description
What do you want to change?
Let online runs price from the cached LiteLLM response, the existing ETag cache, when the server confirmed it recently. A run would go to the network only when that cache is stale or has no entry for a model the report needs:
- Confirmed within the last hour: no request.
- Confirmed within the last day: refresh once, on the first model the cache cannot price exactly.
- Older or missing: refresh up front, as today.
--offline stays as it is.
Why?
Every online run still waits on a LiteLLM round trip, and a unified ccusage daily makes 4 of them. On my machine that adds about 0.4 to 0.7 seconds to each run. This is the same ask as #1412 and #1627.
How? (optional)
I have it implemented and tested on my fork: main...liuzemei:ccusage:perf/cache-first-litellm-pricing
On my own Claude Code and Codex logs, per-model costs match an online run exactly. I'd like to open it as a PR if this direction works for you.
Activity
pullfrog commentedon Oct 25, 2026
github-actions commentedon Oct 25, 2026
Pullfrog triage: priority:low
Kind: feature request
Maintenance fit: excluded
Confidence: high
Requests cache-first LiteLLM pricing to save 0.4-0.7s per run. This is a performance enhancement, not a defect, with no maintainer endorsement, so excluded by default. Prior similar issues were auto-closed minutes after filing and remain unimplemented, so this is not a duplicate.
Decision: closed by the contribution gate.