v0.6.4
Features
- Collapse TokensV2 model sections to a per-model headline. Each model section in the TokensV2 breakdown now shows just a headline (cost) by default, with the full detail revealed on expand — keeping the panel scannable when many models are present. #294
- Exclude sub-cent (
< $0.01) sessions from the Cost Distribution card.$0/tiny/negative/unpriced data points are now dropped before bucketing — they no longer appear in any band, the percentiles, or the covered count. The"< $0.01"floor band is gone; the histogram starts at the$0.01decade. #292
Bug Fixes
- Add gpt-5.4 pricing to
pricing.json. The base model entry was missing, so gpt-5.4 sessions were unpriced. #293
DB Migrations
None.
API Changes
The cost_distribution card in the analytics response (GET /api/v1/analytics) changed shape (3tr4, #292). No routes added or removed; this is a response-schema change to an existing endpoint:
buckets[]no longer includes the"< $0.01"floor band. Bands now start at the$0.01decade (loof the first band is0.01). A range with no priced sessions yields[].stats(p50/p90/p99/avg) are computed only over data points priced>= $0.01;nullwhen there are no priced data points.covered_session_countcounts only sessions priced>= $0.01, so it is ≤ the count of sessions that merely carrytokens_v2data.- Caption copy updated to "Covers N of M sessions priced ≥ $0.01".
See backend/API.md for the full field-by-field documentation.
Breaking Changes
None. No renamed/removed env vars and no CLI impact. The cost_distribution schema change above is additive-safe for tolerant readers (a band was removed and counts narrowed, not a field renamed).