Multiple-choice Question Answering on MMLU Pro (Subject Accuracy)
82.8Biology AccuracyGRPO
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GRPOModel Scale=Qwen2.5-14B-Instruct, Training steps=4002026.04 | 82.8 | 79 | 63.4 | 72.5 | 73.4 | 52.5 | 69.7 | 70.5 | |
| AlphaRLModel Scale=Qwen2.5-14B-Instruct, Training steps=2502026.04 | 82.8 | 76.7 | 63.9 | 70.7 | 72.9 | 51.5 | 66.8 | 69.3 | |
| NExtModel Scale=Qwen2.5-14B-Instruct, Training steps=2502026.04 | 82.6 | 78.2 | 64.3 | 73 | 73.2 | 52.7 | 69.5 | 70.5 | |
| RL-ExtraModel Scale=Qwen2.5-14B-Instruct, Training steps=2502026.04 | 81.1 | 80 | 63.3 | 69.8 | 73 | 50.6 | 68.8 | 69.5 | |
| BackboneModel Scale=Qwen2.5-14B-Instruct2026.04 | 80.4 | 72 | 61.1 | 63.7 | 70.9 | 48.1 | 66.8 | 66.1 | |
| BackboneModel Scale=Qwen2.5-7B-Instruct2026.04 | 77.3 | 65 | 48.4 | 50.3 | 63.2 | 35.3 | 59.1 | 56.9 | |
| NExtModel Scale=Qwen2.5-7B-Instruct, Training steps=2502026.04 | 76.9 | 66.8 | 49.5 | 53.6 | 68.1 | 37.2 | 60 | 58.9 | |
| GRPOModel Scale=Qwen2.5-7B-Instruct, Training steps=4002026.04 | 76.7 | 65.5 | 48.2 | 52.1 | 67.6 | 39.5 | 59.1 | 58.4 | |
| RL-ExtraModel Scale=Qwen2.5-7B-Instruct, Training steps=2502026.04 | 76.7 | 63.7 | 50.3 | 52.4 | 65.1 | 36.7 | 59.9 | 57.8 | |
| AlphaRLModel Scale=Qwen2.5-7B-Instruct, Training steps=2502026.04 | 76.6 | 66.4 | 50.2 | 53.7 | 66.9 | 32.4 | 61.4 | 58.2 | |
| GRPOModel Scale=Qwen2.5-3B-Instruct, Training steps=4002026.04 | 73.6 | 54.4 | 41.9 | 41.5 | 51 | 27 | 48 | 48.2 | |
| NExtModel Scale=Qwen2.5-3B-Instruct, Training steps=2502026.04 | 72.9 | 54.7 | 42.6 | 43.8 | 50.5 | 27.7 | 49.1 | 48.8 | |
| RL-ExtraModel Scale=Qwen2.5-3B-Instruct, Training steps=2502026.04 | 70.3 | 51.4 | 37.4 | 42.7 | 49.1 | 30.2 | 42.8 | 46.3 | |
| AlphaRLModel Scale=Qwen2.5-3B-Instruct, Training steps=2502026.04 | 67.7 | 53.7 | 41.8 | 40.9 | 50.3 | 26.8 | 45.1 | 46.6 | |
| BackboneModel Scale=Qwen2.5-3B-Instruct2026.04 | 65.3 | 49.6 | 35.7 | 39 | 49.2 | 25.3 | 43.7 | 44 | |
| GRPOModel Scale=Qwen2.5-1.5B-Instruct, Training steps=4002026.04 | 48.3 | 39.8 | 22.1 | 27.4 | 40.5 | 13.2 | 26.4 | 31.1 | |
| NExtModel Scale=Qwen2.5-1.5B-Instruct, Training steps=2502026.04 | 46.8 | 40.1 | 22.7 | 28.7 | 39.8 | 13.2 | 25.7 | 31 | |
| AlphaRLModel Scale=Qwen2.5-1.5B-Instruct, Training steps=2502026.04 | 44.2 | 38.2 | 21.7 | 21.1 | 41.7 | 15.9 | 27.9 | 30.1 | |
| RL-ExtraModel Scale=Qwen2.5-1.5B-Instruct, Training steps=2502026.04 | 42.1 | 35.1 | 20.5 | 23.8 | 42.1 | 10.2 | 24.1 | 28.3 | |
| BackboneModel Scale=Qwen2.5-1.5B-Instruct2026.04 | 33.3 | 23 | 14.9 | 14.6 | 29.8 | 11.6 | 19.1 | 20.9 |