Code Generation on HumanEval (Acc and Cost)
86.89AccuracyMACA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MACABackbone=Qwen2.5-14B-Instruct2026.05 | 86.89 | 2,848.9 | |
| MaASBackbone=Qwen2.5-14B-Instruct2026.05 | 85.98 | 3,625.7 | |
| DyLANBackbone=Qwen2.5-14B-Instruct2026.05 | 84.37 | 12,784.6 | |
| AgentPruneBackbone=Qwen2.5-14B-Instruct2026.05 | 82.76 | 3,168.2 | |
| CoTBackbone=Qwen2.5-14B-Instruct2026.05 | 82.15 | 317.5 | |
| VanillaBackbone=Qwen2.5-14B-Instruct2026.05 | 80.54 | 251.3 |