Reward Modeling on JudgeBench (test)
82OverallQwen3-30B-A3B
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Qwen3-30B-A3BSupervision Signal=Outcome and Rationale Supervision2026.02 | 82 | 73.4 | 89.8 | 82.1 | 95.2 | 84.6 | |
| Principles-Qwen32BModel Category=Generative Reward Model2026.02 | 81.4 | 74.6 | 85.7 | 85.7 | 90.5 | 83.8 | |
| GRAM-R²Model Category=Generative Reward Model2026.02 | 81 | 90.9 | 83.7 | 87.5 | 61.9 | 83.4 | |
| Qwen3-14BSupervision Signal=Outcome and Rationale Supervision2026.02 | 79.1 | 66.9 | 86.7 | 91.1 | 90.5 | 82.9 | |
| DeepSeek-R1-0528Model Category=LLM-as-a-Judge2026.02 | 78.8 | 59.1 | 82.7 | 80.4 | 92.9 | 75.8 | |
| RM-R1-Distilled-Qwen-32BModel Category=Generative Reward Model2026.02 | 78.8 | 76 | 80.6 | 88.1 | 70.5 | 81.4 | |
| RM-R1-Distilled-Qwen-14BModel Category=Generative Reward Model2026.02 | 78.1 | 68.1 | 72.4 | 87.8 | 84.2 | 79.8 | |
| Nemotron-SuperModel Category=Generative Reward Model2026.02 | 77.2 | 71.4 | 73.5 | 87.5 | 76.2 | 80 | |
| CE-RM-4BRM Type=Pointwise, Scaling=42026.01 | 76.3 | — | — | — | — | — | |
| RRM-32BModel Category=Generative Reward Model2026.02 | 75.7 | 79.9 | 70.4 | 87.5 | 65 | 74.4 | |
| Qwen3-30B-A3B (Outcome-Only)Supervision Signal=Outcome-Only2026.02 | 75.7 | 65.6 | 87.8 | 82.1 | 76.2 | 80.3 | |
| CE-RM-4BRM Type=Pointwise, Scaling=22026.01 | 75.7 | — | — | — | — | — | |
| RRM-32BRM Type=Pairwise2026.01 | 75.1 | — | — | — | — | — | |
| CE-RM-4BRM Type=Pointwise, Scaling=12026.01 | 73.7 | — | — | — | — | — | |
| TIR-Judge-Zero-4BRM Type=Pointwise2026.01 | 70.4 | — | — | — | — | — | |
| Qwen3-14B (Outcome-Only)Supervision Signal=Outcome-Only2026.02 | 70 | 55.8 | 80.6 | 78.6 | 85.7 | 76.8 | |
| RM-R1-Qwen-32BRM Type=Pairwise2026.01 | 68.6 | — | — | — | — | — | |
| TIR-Judge-Distill-8BRM Type=Pointwise2026.01 | 68.2 | — | — | — | — | — | |
| TIR-Judge-Zero-8BRM Type=Pointwise2026.01 | 67.5 | — | — | — | — | — | |
| Gemini-2.5-FlashRM Type=Pointwise2026.01 | 66.9 | — | — | — | — | — | |
| TIR-Judge-Distill-4BRM Type=Pointwise2026.01 | 66.7 | — | — | — | — | — | |
| RISE-Judge-32BRM Type=Pairwise2026.01 | 66.3 | — | — | — | — | — | |
| CompassJudger2-32BRM Type=Pairwise2026.01 | 65.5 | — | — | — | — | — | |
| Skywork-Reward-Gemma-2-27BModel Category=Scalar Reward Model2026.02 | 65 | 59.7 | 66.3 | 83.9 | 50 | 67.8 | |
| Claude-3.5-SonnetModel Category=LLM-as-a-Judge2026.02 | 64.8 | 62.3 | 66.3 | 66.1 | 64.3 | 62.9 | |
| CompassJudger2-7BRM Type=Pairwise2026.01 | 63.1 | — | — | — | — | — | |
| RewardAnything-8B-v1Model Category=Generative Reward Model2026.02 | 62.6 | 61 | 57.1 | 73.2 | 66.7 | 72.9 | |
| Skywork-Reward-Llama-3.1-8BModel Category=Scalar Reward Model2026.02 | 62.5 | 59.1 | 64.3 | 76.8 | 50 | 66.3 | |
| RRM-7BRM Type=Pairwise2026.01 | 60.9 | — | — | — | — | — | |
| J1-Llama-70BRM Type=Pairwise2026.01 | 60 | — | — | — | — | — | |
| J1-Llama-70BRM Type=Pointwise2026.01 | 60 | — | — | — | — | — | |
| GPT-4oModel Category=LLM-as-a-Judge2026.02 | 59.8 | 50.6 | 54.1 | 75 | 59.5 | 66.2 | |
| RISE-Judge-7BRM Type=Pairwise2026.01 | 59.7 | — | — | — | — | — | |
| RM-R1-Qwen-7BRM Type=Pairwise2026.01 | 59.1 | — | — | — | — | — | |
| J1-Llama-8BRM Type=Pointwise2026.01 | 57.7 | — | — | — | — | — | |
| GPT-4oRM Type=Pairwise2026.01 | 56.6 | — | — | — | — | — | |
| JudgeLRM-7BRM Type=Pairwise2026.01 | 54.6 | — | — | — | — | — | |
| Llama-3.3-70B-InstructRM Type=Pairwise2026.01 | 48.6 | — | — | — | — | — | |
| CompassJudger1-32BRM Type=Pointwise2026.01 | 42.3 | — | — | — | — | — | |
| J1-Llama-8BRM Type=Pairwise2026.01 | 42 | — | — | — | — | — |