ResearchBenchmarksCode Generation on HumanEval (Acc, Avg Cost)Code Generation on HumanEval (Acc, Avg Cost)Follow88.89AccuracyMACA78.375681.105383.83586.5647May 25, 2026Evaluation resultsMethodMethodLinksAccuracyAverage CostMACABackbone=Llama-3.1-70BBackbone=Llama-3.1-70B2026.0588.892,672.2DyLANBackbone=Llama-3.1-70BBackbone=Llama-3.1-70B2026.0584.2512,984.5MaASBackbone=Llama-3.1-70BBackbone=Llama-3.1-70B2026.0583.333,758.5CoTBackbone=Llama-3.1-70BBackbone=Llama-3.1-70B2026.0580.3247.1AgentPruneBackbone=Llama-3.1-70BBackbone=Llama-3.1-70B2026.0580.253,186.9VanillaBackbone=Llama-3.1-70BBackbone=Llama-3.1-70B2026.0578.78239.8