Instruction Following on InfoBench
85.2AccuracyAMARIS
Evaluation results
| Method | Links | |
|---|---|---|
| AMARISRubric Strategy=per-instance rubric2026.05 | 85.2 | |
| AMARISRubric Strategy=global rubric2026.05 | 84.8 | |
| RLCFRL Training Approach=RLCF2026.05 | 84.1 | |
| Rubric-ARMRL Training Approach=Rubric-ARM2026.05 | 83.7 | |
| RubricHub (RuRL)RL Training Approach=RuRL2026.05 | 83.5 | |
| RuscaRLRL Training Approach=RuscaRL2026.05 | 83.2 | |
| OpenRubricsRL Training Approach=DPO via Rubric-RM2026.05 | 83 | |
| Naive (no RL)RL Training Approach=None2026.05 | 78.1 |