Instruction Following on ArenaHard
76.5Hard-Prompt Avg ScoreQUBRIC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| QUBRIC2026.06 | 76.5 | 58.9 | |
| RM-RLHFlength_penalty=1.02026.06 | 71.4 | 42.3 | |
| SFT Baseline2026.06 | 71 | 51.6 | |
| Rubric-RLquery=Original Query2026.06 | 68.6 | 51.2 | |
| AutoIF-RLVR2026.06 | 63.8 | 42.5 | |
| RM-RLHFlength_penalty=none2026.06 | 25.8 | 21.4 |