arazu · Kimi K3 run 1
35.4Library score
Pass rate69.3%
Simplicity65.3
Problems $$14.77
Library $$13.90
The library
arazu · designed by Kimi K3 (mini-SWE)7,407 source lines
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE71.3%59.0$4.24
GLM 5.3 Flash · mini-SWE72.0%67.6$7.07
GPT-5.6 Luna · Codex64.5%69.3$3.45
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
lander-lidar39.849.932.740.8scene-contact34.454.950.046.4world-contact72.279.867.573.2aircraft-raycast48.356.047.550.6capsule-training4.65.53.44.5physics-tick15.326.715.419.1offset-strike47.950.653.850.8layered-laser46.463.957.656.0swept-cargo1.555.435.130.7compound-scanner39.659.765.855.0sensor-transit-ledger43.635.560.146.4articulated-creature0.50.60.40.5cable-replay11.00.00.03.7damping24.329.637.830.6one-way39.528.523.630.5material-braking29.229.723.427.4platformer-ccd37.456.90.031.4grab-replay29.626.641.932.7keyed-platform-drop63.535.937.645.6ccd-wall34.130.639.434.7motor-target35.218.00.017.8rope-cut30.557.165.951.2