P
npr · Production library run 3
50.5Library score
Pass rate88.0%
Simplicity60.6
Problems $$9.20
Implementers use the real production library (ndarray).
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE84.8%58.3$1.98
GLM 5.3 Flash · mini-SWE92.9%59.7$5.64
GPT-5.6 Luna · Codex86.1%63.8$1.58
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
cluster-consensus25.713.721.920.4logistic-path50.963.395.469.9relu-regions13.09.316.612.9table-reasoning68.831.992.564.4embedding-pooling63.066.856.862.2kv-cache67.669.660.866.0channel-stft59.884.158.267.4irls-path72.265.662.966.9peak-covariance0.033.024.319.1envelope-curvature50.956.864.557.4radiance-moments67.162.363.164.2kalman-update84.784.780.983.4csi-pose58.10.00.019.4vmc-update0.180.40.226.9pose-correction83.466.56.952.3spectral-reconstruction39.150.139.042.8portfolio-risk81.077.963.574.1contact-forces60.061.175.865.6density-propagation31.226.956.638.2tensor-train0.053.259.737.6