Closed Question Answering on JAMA (test)
57.8AccuracyGRPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GRPO2026.02 | 57.8 | 2,634 | |
| ESTAR-FTmode=fine-tuned2026.02 | 57.2 | 689 | |
| ESTAR-LITEearly_stopping=classifier-based2026.02 | 56.6 | 419 | |
| ESTARoptimization=RL2026.02 | 56.1 | 352 | |
| FlashThink2026.02 | 55.8 | 2,161 | |
| AdaptThink2026.02 | 55.8 | 1,102 | |
| Length-Penaltyexplicit_length_penalty=true2026.02 | 55.4 | 1,872 | |
| O1-Pruner2026.02 | 53.2 | 1,981 | |
| No-Thinkingreasoning=disabled2026.02 | 48.2 | 369 |