Skip to content
Hbench

meta-harness

meta-harness

meta-harness's pass rate on every board it ran on. Scores on different boards don't add up, so this page doesn't add them.

On each board

Terminal-Bench 2.0

terminal-bench@2.0
Claude Opus 4.61.1.0, full, 89 tasks, 5 attempts, open
Ranks 1–4
8 harnesses
76.4%69.2–83.4meta-harness minus Terminus 2 2.0.0: +13.5 points, 95% interval +7.2 to +20.7, over 89 tasks both ran; a clear difference.

How meta-harness works

Each hexagon shows how meta-harness works, ranked among the harnesses on that board: outward is better.

76.4%pass rate, 95% interval 69.2–83.4

meta-harness 1.1.0, Lean contextmeta-harness 1.1.0, Cache usemeta-harness 1.1.0, Tests before finishingmeta-harness 1.1.0, Direct pathmeta-harness 1.1.0, No loopsmeta-harness 1.1.0, Speed

Compare with another harness

meta-harness 1.1.0, Lean contextmeta-harness 1.1.0, Cache usemeta-harness 1.1.0, Tests before finishingmeta-harness 1.1.0, Direct pathmeta-harness 1.1.0, No loopsmeta-harness 1.1.0, SpeedJudy 0.7.0, Lean contextJudy 0.7.0, Tests before finishingJudy 0.7.0, Direct pathJudy 0.7.0, No loopsJudy 0.7.0, Speed
  • meta-harness 1.1.0 · 76.4%
  • Judy 0.7.0 · 71.9%