weft · Kimi K3 run 2
37.1Library score
Pass rate83.5%
Simplicity51.1
Problems $$10.37
Library $$13.20
The library
weft · designed by Kimi K3 (mini-SWE)4,164 source lines
ImplementerPass rateSimplicityCostScore
DeepSeek V4.1 Flash · mini-SWE88.2%49.9$2.14
GLM 5.3 Flash · mini-SWE76.3%55.5$5.83
GPT-5.6 Luna · Codex85.4%48.3$2.40
ProblemDeepSeek V4.1 FlashGLM 5.3 FlashGPT-5.6 LunaMean
bag-rules32.330.931.331.5ninja-build31.724.737.331.3mail-addresses37.834.256.342.7property-paths39.822.260.440.8formula-ledger31.346.044.240.5record-spans68.564.643.759.0block-events45.662.643.950.7command-errors29.336.318.628.1route-tags62.335.043.647.0template-audit36.20.020.118.8grammar-inventory31.924.836.531.1grammar-trace57.165.941.154.7nesting-guard19.30.08.09.1unicode-spans32.337.133.334.3