General QA on MMLU-Redux
90.89Exact MatchSnapMLA
Evaluation results
| Method | Links | |
|---|---|---|
| SnapMLABackbone=DeepSeek-V3.1, Precision=FP82026.02 | 90.89 | |
| FlashMLABackbone=DeepSeek-V3.1, Precision=BF162026.02 | 90.48 | |
| FlashMLABackbone=LongCat-Flash-thinking, Precision=BF162026.02 | 89.3 | |
| SnapMLABackbone=LongCat-Flash-thinking, Precision=FP82026.02 | 88.27 | |
| DeepSeek-R12025.12 | 88.17 | |
| CompassMax-V3-Thinking2025.12 | 87.23 | |
| CompassMax-V32025.12 | 42.87 |