Visual Equivalence Reward Modeling on Visual-ERM-Bench SVG
33.3F1h ScoreGemini-3-Flash
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Gemini-3-FlashCategory=Proprietary2026.03 | 33.3 | 67.5 | 64.3 | |
| Gemini-2.5-ProCategory=Proprietary2026.03 | 29.3 | 34.3 | 63.3 | |
| GPT-5.2Category=Proprietary2026.03 | 28.5 | 32.2 | 61.1 | |
| Visual-ERMCategory=Open-source, Backbone=Qwen3-VL-8B-Instruct2026.03 | 28.3 | 32.6 | 59.6 | |
| Qwen3-VL-235B-InstructCategory=Open-source2026.03 | 19.4 | 22.8 | 51.5 | |
| GPT-4oCategory=Proprietary2026.03 | 13 | 19.3 | 50.3 | |
| InternVL3.5-8BCategory=Open-source2026.03 | 6.1 | 13.1 | 48.9 | |
| Qwen3-VL-8B-InstructCategory=Open-source2026.03 | 6.1 | 9.4 | 27.1 | |
| Qwen2.5-VL-7BCategory=Open-source2026.03 | 1.9 | 7.5 | 37.9 |