Language Understanding on MMLU-Pro OOD (test)
80.92Biology Avg@16CAST
Evaluation Results
| Method | Links | ||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CASTBackbone=Qwen3-4B, Generation token budget=8K2026.05 | 80.92 | 94.28 | 73.72 | 89.73 | 75.43 | 93.11 | 70.9 | 88.78 | 73.88 | 89.45 | 55.13 | 89.47 | 59.59 | 79.34 | 48.16 | 70.08 | 30.81 | 64.85 | 85.21 | 95.19 | 55.1 | 76.52 | 53.08 | 75.35 | 75.27 | 92.3 | 66.72 | 82.33 | 65.74 | 85.43 | |
| RLRTBackbone=Qwen3-4B, Generation token budget=8K2026.05 | 79.38 | 93.03 | 70.54 | 90.11 | 71.36 | 91.43 | 68.99 | 87.56 | 71.9 | 87.56 | 53.77 | 87.62 | 57.94 | 78.85 | 47.77 | 70.34 | 29.53 | 62.49 | 81.3 | 94.3 | 53.45 | 76.62 | 51.19 | 72.95 | 71.43 | 91.69 | 66.17 | 81.08 | 63.41 | 84.32 | |
| GRPOBackbone=Qwen3-4B, Generation token budget=8K2026.05 | 79.31 | 93.03 | 70.32 | 88.97 | 70.56 | 91.17 | 68.63 | 88.78 | 72.03 | 88.98 | 53.59 | 88.34 | 57.93 | 79.58 | 47.28 | 70.87 | 29.62 | 64.03 | 80.83 | 93.93 | 53.27 | 74.89 | 51.01 | 73.35 | 71.21 | 91.76 | 66.03 | 81.7 | 63.18 | 84.52 | |
| RLSDBackbone=Qwen3-4B, Generation token budget=8K2026.05 | 79.15 | 93.03 | 70.56 | 89.35 | 70.76 | 92.4 | 68.57 | 87.56 | 72.04 | 88.51 | 54.37 | 88.24 | 57.64 | 78.85 | 47.47 | 70.08 | 29.58 | 62.31 | 80.75 | 94.45 | 53.38 | 76.08 | 50.56 | 72.55 | 70.78 | 91.69 | 65.88 | 81.08 | 63.17 | 84.41 | |
| BaseBackbone=Qwen3-4B, Generation token budget=8K2026.05 | 79.02 | 93.03 | 70.08 | 89.48 | 70.56 | 92.14 | 67.79 | 88.78 | 72.07 | 88.39 | 54.35 | 87.72 | 58.21 | 78 | 47.85 | 68.77 | 29.67 | 62.4 | 80.46 | 94.08 | 53.34 | 75.87 | 51.58 | 74.35 | 70.78 | 91.61 | 65.97 | 82.08 | 63.16 | 84.37 | |
| OPSDBackbone=Qwen3-4B, Generation token budget=8K2026.05 | 78.71 | 92.61 | 70.66 | 88.21 | 70.43 | 91.61 | 68.05 | 87.56 | 71.91 | 88.27 | 53.93 | 87.72 | 57.84 | 78.61 | 47.9 | 69.55 | 29.79 | 62.58 | 80.67 | 94.3 | 53.24 | 75.43 | 50.74 | 73.55 | 70.98 | 91.61 | 66.15 | 81.33 | 63.13 | 84.15 |