Multiple-choice Question Answering on MMLU-Pro Target Reasked Group (test)
77.78Accuracy (R1)Qwen3 235B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3 235BStatic Entropy Threshold (tau_H)=1.3 nats, Static Max Probability Threshold (tau_MSP)=0.4, Inference Protocol=Reinforcement Inference2026.02 | 77.78 | 79.53 | |
| Qwen3 30BStatic Entropy Threshold (tau_H)=1.3 nats, Static Max Probability Threshold (tau_MSP)=0.4, Inference Protocol=Reinforcement Inference2026.02 | 73.41 | 74.94 | |
| DeepSeek V3.2Static Entropy Threshold (tau_H)=1.3 nats, Static Max Probability Threshold (tau_MSP)=0.4, Inference Protocol=Reinforcement Inference2026.02 | 31.68 | 44.61 |