General Language Capability on MMLU, GSM8K, and GPQA
73.6MMLU AccuracyMUSE-D
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MUSE-DBackbone=Qwen2.5-7B-IT2026.05 | 73.6 | 82.5 | 32.8 | |
| SafeMTBackbone=Qwen2.5-7B-IT2026.05 | 73.5 | 81 | 32.6 | |
| TRACEBackbone=Qwen2.5-7B-IT, variant=w/o lat.2026.05 | 73.5 | 83.5 | 32 | |
| TRACEBackbone=Qwen2.5-7B-IT, variant=ours2026.05 | 73.5 | 84.5 | 35.6 | |
| Qwen2.5-7B-IT2026.05 | 73.3 | 86.5 | 34.2 |