Question Answering on GPQA No-think (test)
47.8AccuracyQwen3-8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-8BTraining=RL, Mode=No-Think2026.01 | 47.8 | 4,435 | 9.863 | |
| Qwen3-4BTraining=RL, Mode=No-Think2026.01 | 40.4 | 6,871 | 17.196 | |
| Qwen3-4BTraining=RL, Mode=Mid-Think2026.01 | 38.2 | 2,086 | 4.79 | |
| Qwen3-8BTraining=No, Mode=Untrained2026.01 | 37.5 | 1,335 | 0.459 | |
| Qwen3-4BTraining=No, Mode=Untrained2026.01 | 36.9 | 1,560 | 0.59 | |
| Qwen3-8BTraining=RL, Mode=Mid-Think2026.01 | 36.2 | 1,293 | 1.968 | |
| Qwen3-4BTraining=RL, Mode=Think2026.01 | 35.5 | 1,789 | 0.252 | |
| Qwen3-8BTraining=RL, Mode=Think2026.01 | 34.7 | 1,567 | 2.179 |