Multi-task Language Understanding on MMLU STEM
89.1AccuracyMiMo-V2 Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| MiMo-V2 FlashShots=5-shot, # Total Params=309B, # Active Params=15.1B2026.05 | 89.1 | |
| Qwen3.5Shots=5-shot, # Total Params=35B, # Active Params=2.6B2026.05 | 80.2 | |
| LAGUNA XS.2Shots=5-shot, # Total Params=33.4B, # Active Params=3B2026.05 | 78.1 | |
| Nemotron-3-NanoShots=5-shot, # Total Params=31.6B, # Active Params=3.6B2026.05 | 75.3 | |
| Gemma-4Shots=5-shot, # Total Params=25.2B, # Active Params=3.8B2026.05 | 70.4 | |
| LMNetBackbone=Qwen2.5-0.5B2025.05 | 46 | |
| Self-ConsistencyBackbone=Qwen2.5-0.5B, Samples=162025.05 | 41.4 | |
| Self-RefineBackbone=Qwen2.5-0.5B, Passes=162025.05 | 40.6 | |
| PromptBackbone=Qwen2.5-0.5B2025.05 | 39.8 | |
| SFTBackbone=Qwen2.5-0.5B2025.05 | 39.8 | |
| PolychromaticLMstage=SFT, shot=5-shot2026.03 | 28.42 | |
| PolychromaticLMstage=Base, shot=5-shot2026.03 | 25.28 | |
| Randomshot=5-shot2026.03 | 25 |