Multitask Language Understanding on MMLU Pro (Acc, Cost)
64.46AccuracyMACA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MACABackbone=Qwen2.5-14B-Instruct2026.05 | 64.46 | 2,412.5 | |
| DyLANBackbone=Qwen2.5-14B-Instruct2026.05 | 64.08 | 11,960.2 | |
| AgentPruneBackbone=Qwen2.5-14B-Instruct2026.05 | 63.71 | 2,958.4 | |
| MaASBackbone=Qwen2.5-14B-Instruct2026.05 | 63.22 | 3,147.9 | |
| CoTBackbone=Qwen2.5-14B-Instruct2026.05 | 60.94 | 286.7 | |
| VanillaBackbone=Qwen2.5-14B-Instruct2026.05 | 59.12 | 248.6 |