Repeatable test setup

TokenWarden used up to 70% fewer input tokens.

We ran the same opencode coding tasks with TokenWarden, with no plugin, and with other token-saving plugins.

The full suite includes 210 runs and 7.7 hours of runtime. The detailed setup and raw results are available in the benchmark repository.

Latest comparison

Average input tokens used.

These results use the 30-run read/write subset behind the up-to-70% claim. Percent saved is measured against the no-plugin result.

SetupRunsAverage tokensAverage savings
No plugins6336,849Baseline
TokenWarden6101,21370% saved
OpenSlimEdit6393,09216.7% higher usage
DCP6261,27722.4% saved
OpenRTK6325,1053.5% saved