arazu · DeepSeek V4 Pro run 1

19.1Library score
Pass rate72.5%
Simplicity39.9
Problems $$16.47
Library $$0.30
The library
arazu · designed by DeepSeek V4 Pro (mini-SWE)3,730 source lines
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE71.0%38.2$3.20
18.6
GLM 5.3 Flash · mini-SWE77.2%34.2$9.62
20.0
GPT-5.6 Luna · Codex69.0%47.8$3.65
18.9
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
lander-lidar28.127.726.227.3scene-contact4.816.413.311.5world-contact27.335.10.020.8aircraft-raycast23.45.810.213.1capsule-training3.17.16.45.5physics-tick9.74.63.66.0offset-strike15.216.320.417.3layered-laser37.448.845.844.0swept-cargo0.012.312.08.1compound-scanner15.73.425.314.8sensor-transit-ledger5.648.534.829.6articulated-creature11.80.30.64.2cable-replay0.016.323.213.2damping24.217.527.823.2one-way27.631.813.124.2material-braking19.017.022.519.5platformer-ccd29.532.326.129.3grab-replay24.519.125.022.9keyed-platform-drop31.427.530.429.8ccd-wall21.014.311.715.7motor-target19.410.40.09.9rope-cut29.926.536.831.1