P

npr · Production library run 1

46.4Library score

Pass rate84.9%

Simplicity59.2

Problems $$10.76

Implementers use the real production library (ndarray).

ImplementerPass rateSimplicityCostScore

DeepSeek V4.1 Flash · mini-SWE85.1%55.3$2.10

43.6

GLM 5.3 Flash · mini-SWE80.8%62.9$7.00

45.5

GPT-5.6 Luna · Codex88.6%59.4$1.67

50.2

ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean

cluster-consensus13.914.631.720.0 logistic-path38.541.877.152.5 relu-regions12.022.522.519.0 table-reasoning34.954.451.947.1 embedding-pooling68.281.958.669.5 kv-cache0.10.460.420.3 channel-stft49.355.453.352.7 irls-path66.858.557.861.0 peak-covariance1.540.827.323.2 envelope-curvature48.159.157.855.0 radiance-moments73.945.058.259.1 kalman-update81.182.882.882.2 csi-pose59.10.00.019.7 vmc-update67.70.20.222.7 pose-correction75.577.873.275.5 spectral-reconstruction38.09.541.929.8 portfolio-risk71.486.668.575.5 contact-forces39.264.250.851.4 density-propagation32.733.362.842.9 tensor-train0.080.567.549.3