arazu · GLM 5.3 run 3

28.9Library score
Pass rate66.2%
Simplicity62.9
Problems $$17.54
Library $$21.05
The library
arazu · designed by GLM 5.3 (mini-SWE)7,337 source lines
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE66.1%58.3$2.56
26.9
GLM 5.3 Flash · mini-SWE65.9%63.1$11.15
28.8
GPT-5.6 Luna · Codex66.6%67.4$3.83
31.1
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
lander-lidar34.452.645.644.2scene-contact25.723.717.522.3world-contact34.133.729.832.5aircraft-raycast0.032.50.010.8capsule-training4.75.65.75.3physics-tick12.77.313.711.2offset-strike27.428.630.328.8layered-laser54.472.470.865.9swept-cargo1.113.127.613.9compound-scanner33.352.649.645.2sensor-transit-ledger38.149.849.945.9articulated-creature28.30.40.69.8cable-replay29.70.036.622.1damping25.120.729.124.9one-way36.727.342.435.5material-braking28.032.332.430.9platformer-ccd32.338.338.436.3grab-replay38.035.841.938.6keyed-platform-drop44.30.027.023.7ccd-wall15.852.119.029.0motor-target18.70.924.514.7rope-cut28.554.551.244.7