Reward Modeling on RM-Bench Chat
78.5AccuracyGemini-2.5-Flash
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini-2.5-Flash2026.02 | 78.5 | |
| Gemini-2.5-FlashModel Category=Black-box LLMs2026.05 | 78.5 | |
| RM-R1-32B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst, Size=32B2026.02 | 75.3 | |
| RM-R1-32B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst2026.05 | 75.3 | |
| RM-R1-32B (DeepSeek-Dist)Backbone=DeepSeek-Dist, Size=32B2026.02 | 74.2 | |
| RM-R1-32B (DeepSeek-Dist)Backbone=DeepSeek-Dist2026.05 | 74.2 | |
| RRM-32BSize=32B2026.02 | 73.9 | |
| RRM-32BModel Category=Larger White-box LLMs2026.05 | 73.9 | |
| RM-R1-14B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst, Size=14B2026.02 | 73.2 | |
| RM-R1-14B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst2026.05 | 73.2 | |
| RM-R1-14B (DeepSeek-Dist)Backbone=DeepSeek-Dist, Size=14B2026.02 | 71.4 | |
| RM-R1-14B (DeepSeek-Dist)Backbone=DeepSeek-Dist2026.05 | 71.4 | |
| Rubric-ARM2026.02 | 69.2 | |
| Rubric-ARM-voting@5voting=52026.02 | 69.1 | |
| Rubric-ARROW-voting@5Voting Strategy=voting@52026.05 | 68.6 | |
| API (Rubric+Judge)Rubric Model=GPT-4.1-Mini, Judge Model=Gemini-2.5-Flash Lite2026.02 | 67.9 | |
| API Prompting (Rubric+Judge pairwise)Evaluation Protocol=Rubric+Judge pairwise2026.05 | 67.9 | |
| Rubric-ARROWModel Category=Rubric-based Methods2026.05 | 67.8 | |
| API (direct Judge)Judge Model=Gemini-2.5-Flash Lite2026.02 | 67.2 | |
| API Prompting (direct Judge)Evaluation Protocol=direct Judge2026.05 | 67.2 | |
| Rubric-RM-voting@5voting=52026.02 | 67 | |
| Rubric-RM-voting@5Voting Strategy=voting@52026.05 | 67 | |
| Rubric-ARROW w/o RL-voting@5Voting Strategy=voting@5, Training Protocol=w/o RL2026.05 | 66.9 | |
| Rubric-RM2026.02 | 65.7 | |
| Rubric-RMModel Category=Rubric-based Methods2026.05 | 65.7 | |
| Rubric-ARROW w/o RLTraining Protocol=w/o RL2026.05 | 64.8 | |
| RIFLModel Category=Rubric-based Methods2026.05 | 64.8 | |
| RIFL-voting@5Voting Strategy=voting@52026.05 | 64.3 | |
| RM-R1-7B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst, Size=7B2026.02 | 64.2 | |
| Qwen-3-8B (Rubric+Judge)Backbone=Qwen-3-8B2026.02 | 64.2 | |
| RM-R1-7B (Qwen-2.5-Inst)Backbone=Qwen-2.5-Inst2026.05 | 64.2 | |
| Qwen-3-8B (Rubric+Judge pairwise)Backbone=Qwen-3-8B, Evaluation Protocol=Rubric+Judge pairwise2026.05 | 64.2 | |
| Claude-3.5-Sonnet2026.02 | 62.5 | |
| Claude-3.5-SonnetModel Category=Black-box LLMs2026.05 | 62.5 | |
| RM-R1-7B (DeepSeek-Dist)Backbone=DeepSeek-Dist, Size=7B2026.02 | 62.2 | |
| RM-R1-7B (DeepSeek-Dist)Backbone=DeepSeek-Dist2026.05 | 62.2 | |
| Qwen-3-8B (Rubric+Judge pointwise)Backbone=Qwen-3-8B, Evaluation Protocol=Rubric+Judge pointwise2026.05 | 61.2 | |
| RRM-7BSize=7B2026.02 | 59.9 | |
| RRM-7BModel Category=White-box Judge/Reward LLMs2026.05 | 59.9 | |
| JudgeLRM-7BSize=7B2026.02 | 55.4 | |
| JudgeLRM-7BModel Category=White-box Judge/Reward LLMs2026.05 | 55.4 | |
| API Prompting (Rubric+Judge pointwise)Evaluation Protocol=Rubric+Judge pointwise2026.05 | 37.8 |