arazu · Kimi K3 run 2

41.7Library score
Pass rate75.9%
Simplicity67.0
Problems $$13.47
Library $$23.96
The library
arazu · designed by Kimi K3 (mini-SWE)10,601 source lines
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE75.3%63.7$3.72
39.4
GLM 5.3 Flash · mini-SWE71.5%69.3$6.45
43.2
GPT-5.6 Luna · Codex81.3%68.1$3.30
42.3
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
lander-lidar49.350.642.747.5scene-contact42.157.051.450.2world-contact86.588.10.058.2aircraft-raycast52.60.054.535.7capsule-training28.60.224.017.6physics-tick11.714.246.624.2offset-strike47.849.055.350.7layered-laser90.377.890.786.3swept-cargo0.91.829.410.7compound-scanner20.971.748.647.1sensor-transit-ledger41.860.059.153.6articulated-creature31.60.40.610.8cable-replay0.064.041.635.2damping39.551.944.345.2one-way38.927.444.236.8material-braking32.843.732.536.3platformer-ccd30.258.254.247.5grab-replay57.548.641.849.3keyed-platform-drop50.134.539.941.5ccd-wall22.041.333.132.1motor-target35.334.245.438.3rope-cut57.176.451.761.8