发布

  • [NA] [BE][FE] Make AI Spend cost computation model-aware

    frostbyte_neo 发布于 2026-06-12 14:28:23 +00:00

    AI Spend priced an entire window at a single representative model
    (models[0] / anyLast(model)), so workspaces mixing models (e.g. Opus +
    Haiku) could be costed several-fold wrong. Group every tier-bearing
    figure by cc.billing.model and price each model at its own rate.

    • BE: composition, summary, users and breakdown queries now GROUP BY
      model; responses carry a per-model ModelTiers[] (by_model) instead of
      flat tiers + a single model; new SpendSummaryResponse ships per-model
      current/previous spend; breakdown folds the tail into other per
      model; users derive the model from cc.billing instead of spans.
    • FE: tiersCost() prices each model and sums; dominantModel() drives the
      leaderboard chip (+N when a user spans several models).

    Co-Authored-By: Claude Opus 4.8 noreply@anthropic.com

    下载附件