Human Consistency Evaluation on GenAI-Bench
38.4Kendall's Tau-cVQAScore
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VQAScoreMethod Category=LLM-based, Backbone=CLIP-FlanT5-XXL2026.05 | 38.4 | — | 37.9 | 49.9 | |
| EXPERTMethod Category=LLM-based2026.05 | 35.6 | — | 35.1 | 46.3 | |
| BEiTScoreMethod Category=Encoder-based, Backbone=ViT-L/142026.05 | 35.6 | — | 35.1 | 46.5 | |
| FLEURMethod Category=LLM-based2026.05 | 34 | — | 33.5 | 44.5 | |
| BEiTScoreMethod Category=Encoder-based, Backbone=ViT-L/14, Training Stage=w/o Stage 12026.05 | 33.1 | — | 32.7 | 43.2 | |
| BEiTScoreMethod Category=Encoder-based, Backbone=ViT-B/322026.05 | 33.1 | — | 32.7 | 43.2 | |
| BEiTScoreMethod Category=Encoder-based, Backbone=ViT-B/32, Training Stage=w/o Stage 12026.05 | 28 | — | 27.6 | 36.7 | |
| PAC-S++Method Category=Encoder-based, Backbone=ViT-L2026.05 | 16.3 | — | 16.1 | 21.6 | |
| LongCLIP-SMethod Category=Encoder-based, Backbone=ViT-L2026.05 | 15 | — | 14.8 | 19.8 | |
| PAC-S++Method Category=Encoder-based, Backbone=ViT-B2026.05 | 14.6 | — | 14.4 | 19.3 | |
| LongCLIP-SMethod Category=Encoder-based, Backbone=ViT-B2026.05 | 13.4 | — | 13.2 | 17.7 | |
| PAC-SMethod Category=Encoder-based, Backbone=ViT-B2026.05 | 12.9 | — | 12.7 | 17.1 | |
| PAC-SMethod Category=Encoder-based, Backbone=ViT-L2026.05 | 12.2 | — | 12.1 | 16.3 | |
| CLIP-SMethod Category=Encoder-based, Backbone=ViT-L2026.05 | 11.8 | — | 11.6 | 15.6 | |
| CLIP-SMethod Category=Encoder-based, Backbone=ViT-B2026.05 | 10 | — | 9.9 | 13.3 | |
| SPECSMethod Category=Encoder-based2026.05 | 1.5 | — | 1.4 | 1.9 | |
| Gemini-2.5-ProModel Category=Closed-source2025.06 | — | 61.6 | — | — | |
| GPT-4oModel Category=Closed-source2025.06 | — | 53.6 | — | — | |
| LLaVA-CriticModel Category=Open-source, Parameters=7B2025.06 | — | 29.8 | — | — | |
| LLaVA-CriticModel Category=Open-source, Parameters=72B2025.06 | — | 46.8 | — | — | |
| LLaVA-OVModel Category=Open-source, Parameters=7B2025.06 | — | 13.6 | — | — | |
| LLaVA-OVModel Category=Open-source, Parameters=72B2025.06 | — | 46.3 | — | — | |
| MinosModel Category=Open-source, Parameters=8B2025.06 | — | 55.9 | — | — | |
| Prometheus-VModel Category=Open-source, Parameters=7B2025.06 | — | 16 | — | — | |
| Qwen3-VLModel Category=Open-source, Parameters=8B2025.06 | — | 53.6 | — | — | |
| UnifiedReward_LModel Category=Open-source, Parameters=7B2025.06 | — | 54.4 | — | — | |
| UnifiedReward_QModel Category=Open-source, Parameters=8B2025.06 | — | 54.7 | — | — |