arazu · Kimi K3 run 3

34.6Library score
Pass rate71.8%
Simplicity65.1
Problems $$14.02
Library $$20.74
The library
arazu · designed by Kimi K3 (mini-SWE)7,794 source lines
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE79.9%60.1$3.92
36.8
GLM 5.3 Flash · mini-SWE69.6%69.9$6.35
37.0
GPT-5.6 Luna · Codex65.7%65.3$3.76
30.0
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
lander-lidar41.347.235.841.4scene-contact26.138.924.029.7world-contact28.036.50.021.5aircraft-raycast41.627.754.341.2capsule-training26.85.121.417.7physics-tick13.99.018.813.9offset-strike50.578.944.858.1layered-laser77.685.389.084.0swept-cargo45.02.427.625.0compound-scanner25.453.550.143.0sensor-transit-ledger46.256.946.249.8articulated-creature41.80.40.714.3cable-replay30.127.20.019.1damping27.233.41.220.6one-way45.835.257.946.3material-braking33.835.037.435.4platformer-ccd35.648.640.941.7grab-replay49.246.353.449.6keyed-platform-drop53.344.437.745.1ccd-wall25.512.610.516.2motor-target35.413.00.016.1rope-cut10.076.19.331.8