Human Consistency Evaluation on RichHF-18K
33.9Kendall's TauGemini-2.5-Pro
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini-2.5-ProModel Category=Closed-source2025.06 | 33.9 | |
| Qwen3-VLModel Category=Open-source, Parameters=8B2025.06 | 33.7 | |
| UnifiedReward_QModel Category=Open-source, Parameters=8B2025.06 | 33.6 | |
| UnifiedReward_LModel Category=Open-source, Parameters=7B2025.06 | 33.4 | |
| MinosModel Category=Open-source, Parameters=8B2025.06 | 31.5 | |
| LLaVA-CriticModel Category=Open-source, Parameters=72B2025.06 | 28.9 | |
| GPT-4oModel Category=Closed-source2025.06 | 27.1 | |
| LLaVA-OVModel Category=Open-source, Parameters=72B2025.06 | 24.3 | |
| LLaVA-CriticModel Category=Open-source, Parameters=7B2025.06 | 16.8 | |
| Prometheus-VModel Category=Open-source, Parameters=7B2025.06 | 6.55 | |
| LLaVA-OVModel Category=Open-source, Parameters=7B2025.06 | 3.15 |