Your savings number, told honestly and in full
This release reworks how Token Optimizer reports what it saves you, and grounds every number in your own usage.
A true superset headline
Your headline savings now combine every lever in one number: model routing, prompt caching, cache-write, subagents, and the lighter context Token Optimizer keeps for you. It is computed as a current-volume counterfactual, your real 30-day volume held constant and priced two ways, the way you worked before Token Optimizer versus the way you work now. Because the volume is identical on both sides, the gap is pure efficiency.
The directly-metered savings stay reported separately as the proven floor, never summed on top of the estimate.
A baseline that is actually yours
The before/after is frozen from your own earliest sessions, not a generic assumption. A deep first-run backfill reads the history already on your disk, so existing users get an immediate baseline instead of waiting for a fresh 30-day window. While the baseline is still forming, the dashboard shows clear progress rather than a blank state.
Dashboard
- A baseline-building progress state for new installs.
- A prominent panel surfacing additional savings you can still unlock.
- A measured-to-date card labeled by your first tracked session, not install day.
Benchmarks and methodology
The benchmark report and methodology docs now lead with the savings number and are fully reproducible against your own session history. Every figure regenerates from commands that ship in the repo.
Cross-platform
The savings methodology and dashboard shape are carried across to the OpenClaw and OpenCode adapters, with platform differences documented rather than hidden.
Includes the v5.11.9 measurement-accuracy work.