Reward Modeling on AssistantBench
89.17Pairwise AccuracyGPT-4o-mini
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-4o-miniEvaluation Protocol=LLM-as-judge, Model Source=Proprietary2026.01 | 89.17 | 73.33 | |
| WebArbiter-7BEvaluation Protocol=WebPRM, Model Source=WebPRMs, Model Scale=7B2026.01 | 89.17 | 70 | |
| GPT-4oEvaluation Protocol=LLM-as-judge, Model Source=Proprietary2026.01 | 85.83 | 66.67 | |
| Llama-3-70B-InstructEvaluation Protocol=LLM-as-judge, Model Source=Open-source2026.01 | 85.83 | 70 | |
| Qwen2.5-7B-InstructEvaluation Protocol=LLM-as-judge, Model Source=Open-source2026.01 | 84.17 | 53.33 | |
| GPT-5Evaluation Protocol=LLM-as-judge, Model Source=Proprietary2026.01 | 81.67 | 63.33 | |
| Claude-3.7-SonnetEvaluation Protocol=LLM-as-judge, Model Source=Proprietary2026.01 | 81.5 | 61.3 | |
| Gemini-2.5-FlashEvaluation Protocol=LLM-as-judge, Model Source=Proprietary2026.01 | 80 | 63.33 | |
| DeepSeek-R1Evaluation Protocol=LLM-as-judge, Model Source=Proprietary2026.01 | 78.49 | 56.18 | |
| WebArbiter-3BEvaluation Protocol=WebPRM, Model Source=WebPRMs, Model Scale=3B2026.01 | 78.33 | 46.67 | |
| Qwen2.5-3B-InstructEvaluation Protocol=LLM-as-judge, Model Source=Open-source2026.01 | 75.83 | 33.33 | |
| WebShepherd-3BEvaluation Protocol=WebPRM, Model Source=WebPRMs, Model Scale=3B2026.01 | 66.67 | 46.67 | |
| WebShepherd-8BEvaluation Protocol=WebPRM, Model Source=WebPRMs, Model Scale=8B2026.01 | 55.92 | 30 |