arazu · DeepSeek V4 Pro run 3
20.2Library score
Pass rate68.8%
Simplicity41.5
Problems $$14.38
Library $$0.48
The library
arazu · designed by DeepSeek V4 Pro (mini-SWE)1,779 source lines
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE71.8%35.1$2.92
GLM 5.3 Flash · mini-SWE71.4%38.0$8.43
GPT-5.6 Luna · Codex63.4%51.2$3.04
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
lander-lidar24.119.026.923.3scene-contact15.13.017.111.7world-contact30.848.518.232.5aircraft-raycast20.212.425.919.5capsule-training1.53.85.03.4physics-tick7.54.04.65.4offset-strike16.817.511.515.3layered-laser52.518.453.841.6swept-cargo0.40.012.24.2compound-scanner17.523.922.321.2sensor-transit-ledger33.841.847.641.1articulated-creature0.428.80.810.0cable-replay6.00.00.02.0damping17.723.40.713.9one-way27.119.616.621.1material-braking18.218.322.419.6platformer-ccd25.632.536.131.4grab-replay33.429.137.733.4keyed-platform-drop24.034.718.725.8ccd-wall16.813.742.924.5motor-target20.89.90.010.2rope-cut24.635.241.633.8