List your company

Multi-task Language Understanding on MMLU-Pro (Subject Scores)

88.6Biology ScoreDSMoE
64.05670.42876.883.172Apr 7, 2026Apr 20, 2026May 4, 2026May 18, 2026May 31, 2026Jun 14, 2026Jun 28, 2026

Evaluation results

MethodLinks
DSMoEBackbone=Qwen3-30B-Instruct2026.0488.686.38078.383.3—
DSMoEBackbone=GPT-OSS-120B2026.0488.187.582.679.284.4—
SFTBackbone=Qwen3-30B-Instruct2026.0487.985.478.877.182.3—
OriginalBackbone=Qwen3-30B-Instruct2026.0487.68578.176.281.7—
RICEBackbone=Qwen3-30B-Instruct2026.048785.979.177.782.4—
SFTBackbone=GPT-OSS-120B2026.0486.877.480.979.881.2—
RICEBackbone=GPT-OSS-120B2026.0485.882.881.87882.1—
DSMoEBackbone=GPT-OSS-20B2026.0478.774.565.749.667.1—
OriginalBackbone=GPT-OSS-20B2026.0476.466.56248.963.4—
DSMoEBackbone=Qwen3-30B-Thinking2026.0475.278.365.859.369.6—
HIPPOTraining Paradigm=Ours2026.0675————56.8
Qwen2.5_7BTraining Paradigm=Standard Training, Configuration=+ RL2026.0674.5————57.5
RICEBackbone=Qwen3-30B-Thinking2026.0474.374.559.459.366.9—
OriginalBackbone=Qwen3-30B-Thinking2026.0473.975.358.756.566.1—
SFTBackbone=GPT-OSS-20B2026.0473.962.658.247.360.5—
SFTBackbone=Qwen3-30B-Thinking2026.0473.272.861.858.766.6—
RICEBackbone=GPT-OSS-20B2026.0473.164.757.348.961—
OriginalBackbone=GPT-OSS-120B2026.0472.87482.450.269.8—
PREF-GRPOTraining Paradigm=Reward-Shaping2026.0671.7————52.9
SP3FTraining Paradigm=Reward-Shaping2026.0668————55.4
Qwen2.5_7BTraining Paradigm=Standard Training, Configuration=+ SFT2026.0667————48.2
PRM RLTraining Paradigm=Reward-Shaping2026.0665.7————41.1
Qwen2.5_7BTraining Paradigm=Standard Training2026.0665————51.3