P
npr · Production library run 2
46.2Library score
Pass rate83.6%
Simplicity60.5
Problems $$8.82
Implementers use the real production library (ndarray).
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE85.8%62.3$1.80
GLM 5.3 Flash · mini-SWE83.9%59.5$5.48
GPT-5.6 Luna · Codex81.1%59.7$1.55
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
cluster-consensus29.417.216.421.0logistic-path54.939.194.963.0relu-regions18.911.718.416.3table-reasoning100.078.258.779.0embedding-pooling69.272.275.872.4kv-cache60.158.867.362.1channel-stft45.476.848.456.9irls-path60.659.562.060.7peak-covariance25.338.224.929.5envelope-curvature55.755.654.455.2radiance-moments4.960.155.540.2kalman-update84.583.077.481.6csi-pose58.30.00.019.4vmc-update0.10.20.20.2pose-correction70.364.054.863.0spectral-reconstruction42.649.820.337.6portfolio-risk67.268.966.667.6contact-forces93.445.114.651.1density-propagation52.832.059.047.9tensor-train0.00.60.20.3