Visual Equivalence Reward Modeling on Visual-ERM-Bench Chart
39.9F1hVisual-ERM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Visual-ERMCategory=Open-source, Backbone=Qwen3-VL-8B-Instruct2026.03 | 39.9 | 42.8 | 61.2 | |
| Gemini-3-FlashCategory=Proprietary2026.03 | 38.5 | 41.3 | 62.8 | |
| Gemini-2.5-ProCategory=Proprietary2026.03 | 33.7 | 37.5 | 61.8 | |
| GPT-5.2Category=Proprietary2026.03 | 30.1 | 32.6 | 64.8 | |
| Qwen3-VL-235B-InstructCategory=Open-source2026.03 | 28 | 31.8 | 47.2 | |
| GPT-4oCategory=Proprietary2026.03 | 22.8 | 28.3 | 48.5 | |
| Qwen2.5-VL-7BCategory=Open-source2026.03 | 3.9 | 5.4 | 11.2 | |
| Qwen3-VL-8B-InstructCategory=Open-source2026.03 | 3.3 | 3.5 | 3.8 | |
| InternVL3.5-8BCategory=Open-source2026.03 | 2.5 | 5.7 | 11 |