arazu · DeepSeek V4 Pro run 2
18.9Library score
Pass rate73.5%
Simplicity38.5
Problems $$15.64
Library $$0.31
The library
arazu · designed by DeepSeek V4 Pro (mini-SWE)3,524 source lines
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE79.6%32.8$3.63
GLM 5.3 Flash · mini-SWE75.3%34.0$8.52
GPT-5.6 Luna · Codex65.3%49.1$3.49
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
lander-lidar24.420.537.527.5scene-contact14.620.118.517.7world-contact35.130.50.021.9aircraft-raycast19.219.229.122.5capsule-training7.611.08.89.1physics-tick3.19.86.46.4offset-strike15.918.016.916.9layered-laser21.720.032.024.6swept-cargo9.80.014.38.0compound-scanner17.83.122.414.4sensor-transit-ledger21.232.638.630.8articulated-creature28.00.40.39.6cable-replay8.218.60.08.9damping20.621.717.620.0one-way24.922.134.727.2material-braking17.34.315.212.3platformer-ccd26.130.427.628.1grab-replay23.638.410.024.0keyed-platform-drop28.327.822.726.3ccd-wall11.58.635.118.4motor-target30.619.50.016.7rope-cut33.94.133.924.0