–

npr · No library run 2

43.1Library score
Pass rate90.6%
Simplicity49.7
Problems $$4.96

Implementers get no library and write everything from scratch.

ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE98.8%51.1$0.96
49.2
GLM 5.3 Flash · mini-SWE81.7%48.5$3.11
37.0
GPT-5.6 Luna · Codex91.4%49.7$0.89
43.0
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
cluster-consensus33.518.427.626.5logistic-path99.748.652.667.0relu-regions27.810.727.822.1table-reasoning63.642.999.468.6embedding-pooling48.659.143.850.5kv-cache38.846.249.945.0channel-stft55.161.559.658.7irls-path51.652.348.350.7peak-covariance27.942.528.232.9envelope-curvature49.010.143.034.0radiance-moments60.651.457.556.5kalman-update40.937.540.439.6csi-pose45.70.47.317.8vmc-update53.70.00.118.0pose-correction58.161.553.957.8spectral-reconstruction32.831.030.131.3portfolio-risk62.753.945.053.8contact-forces41.837.641.140.2density-propagation41.25.246.531.0tensor-train51.869.157.959.6