Multi-task Language Understanding on MMLU-Pro (Subject Scores)
88.6Biology ScoreDSMoE
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DSMoEBackbone=Qwen3-30B-Instruct2026.04 | 88.6 | 86.3 | 80 | 78.3 | 83.3 | — | |
| DSMoEBackbone=GPT-OSS-120B2026.04 | 88.1 | 87.5 | 82.6 | 79.2 | 84.4 | — | |
| SFTBackbone=Qwen3-30B-Instruct2026.04 | 87.9 | 85.4 | 78.8 | 77.1 | 82.3 | — | |
| OriginalBackbone=Qwen3-30B-Instruct2026.04 | 87.6 | 85 | 78.1 | 76.2 | 81.7 | — | |
| RICEBackbone=Qwen3-30B-Instruct2026.04 | 87 | 85.9 | 79.1 | 77.7 | 82.4 | — | |
| SFTBackbone=GPT-OSS-120B2026.04 | 86.8 | 77.4 | 80.9 | 79.8 | 81.2 | — | |
| RICEBackbone=GPT-OSS-120B2026.04 | 85.8 | 82.8 | 81.8 | 78 | 82.1 | — | |
| DSMoEBackbone=GPT-OSS-20B2026.04 | 78.7 | 74.5 | 65.7 | 49.6 | 67.1 | — | |
| OriginalBackbone=GPT-OSS-20B2026.04 | 76.4 | 66.5 | 62 | 48.9 | 63.4 | — | |
| DSMoEBackbone=Qwen3-30B-Thinking2026.04 | 75.2 | 78.3 | 65.8 | 59.3 | 69.6 | — | |
| HIPPOTraining Paradigm=Ours2026.06 | 75 | — | — | — | — | 56.8 | |
| Qwen2.5_7BTraining Paradigm=Standard Training, Configuration=+ RL2026.06 | 74.5 | — | — | — | — | 57.5 | |
| RICEBackbone=Qwen3-30B-Thinking2026.04 | 74.3 | 74.5 | 59.4 | 59.3 | 66.9 | — | |
| OriginalBackbone=Qwen3-30B-Thinking2026.04 | 73.9 | 75.3 | 58.7 | 56.5 | 66.1 | — | |
| SFTBackbone=GPT-OSS-20B2026.04 | 73.9 | 62.6 | 58.2 | 47.3 | 60.5 | — | |
| SFTBackbone=Qwen3-30B-Thinking2026.04 | 73.2 | 72.8 | 61.8 | 58.7 | 66.6 | — | |
| RICEBackbone=GPT-OSS-20B2026.04 | 73.1 | 64.7 | 57.3 | 48.9 | 61 | — | |
| OriginalBackbone=GPT-OSS-120B2026.04 | 72.8 | 74 | 82.4 | 50.2 | 69.8 | — | |
| PREF-GRPOTraining Paradigm=Reward-Shaping2026.06 | 71.7 | — | — | — | — | 52.9 | |
| SP3FTraining Paradigm=Reward-Shaping2026.06 | 68 | — | — | — | — | 55.4 | |
| Qwen2.5_7BTraining Paradigm=Standard Training, Configuration=+ SFT2026.06 | 67 | — | — | — | — | 48.2 | |
| PRM RLTraining Paradigm=Reward-Shaping2026.06 | 65.7 | — | — | — | — | 41.1 | |
| Qwen2.5_7BTraining Paradigm=Standard Training2026.06 | 65 | — | — | — | — | 51.3 |