Reward Modeling on WEval
95.7CorrelationHuman
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HumanModel Type=Human annotator2026.04 | 95.7 | 97.9 | 79.4 | |
| Our-RM-7BModel Type=Trained reward model2026.04 | 94.6 | 97.3 | 78 | |
| Qwen2.5-72B-Instruct-as-a-judgeModel Type=LLM-as-a-judge2026.04 | 93.6 | 96.8 | 78.7 | |
| Qwen2.5-7B-Instruct-as-a-judgeModel Type=LLM-as-a-judge2026.04 | 85.4 | 92.9 | 60.9 | |
| Skywork-Reward-V2-Qwen3-8BModel Type=Trained reward model, Backbone=Qwen3-8B2026.04 | 82.5 | 91.7 | 45.3 | |
| Writing-Critic-7BModel Type=Trained reward model2026.04 | 80.6 | 90.8 | 42.7 | |
| Skywork-Reward-V2-Llama-3.1-8BModel Type=Trained reward model, Backbone=Llama-3.1-8B2026.04 | 75.4 | 88.7 | 32.9 |