Reward Modeling on FollowBench (Accuracy)
89.9AccuracyRM-R1-14B (DeepSeek-Dist)
Evaluation results
| Method | Links | |
|---|---|---|
| RM-R1-14B (DeepSeek-Dist)Backbone=DeepSeek-Dist, Size=14B2026.02 | 89.9 | |
| RM-R1-32B (DeepSeek-Dist)Backbone=DeepSeek-Dist, Size=32B2026.02 | 89.2 | |
| Rubric-ARM-voting@5voting=52026.02 | 87.4 | |
| Gemini-2.5-Flash2026.02 | 86 | |
| RRM-32BSize=32B2026.02 | 85.7 | |
| Rubric-ARM2026.02 | 85.7 | |
| RM-R1-32B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst, Size=32B2026.02 | 84.9 | |
| RM-R1-14B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst, Size=14B2026.02 | 84 | |
| API (Rubric+Judge)Rubric Model=GPT-4.1-Mini, Judge Model=Gemini-2.5-Flash Lite2026.02 | 83.2 | |
| API (direct Judge)Judge Model=Gemini-2.5-Flash Lite2026.02 | 81.7 | |
| Rubric-RM-voting@5voting=52026.02 | 81.5 | |
| JudgeLRM-7BSize=7B2026.02 | 79.8 | |
| Rubric-RM2026.02 | 76.1 | |
| RM-R1-7B (DeepSeek-Dist)Backbone=DeepSeek-Dist, Size=7B2026.02 | 69.7 | |
| RRM-7BSize=7B2026.02 | 65.5 | |
| Qwen-3-8B (Rubric+Judge)Backbone=Qwen-3-8B2026.02 | 63 | |
| RM-R1-7B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst, Size=7B2026.02 | 56.3 |