Reward Modeling on ARMBench-VL ours (test)
67.6FG ScoreARM-Thinker-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ARM-Thinker-7Bparameters=7B2025.12 | 67.6 | 73.8 | 52.4 | 64.6 | |
| GPT-4o2025.12 | 61.8 | 69.5 | 58.7 | 63.3 | |
| InternVL3-8Bparameters=8B2025.12 | 58.9 | 59.1 | 47 | 55 | |
| InternVL3.5-8Bparameters=8B2025.12 | 56.7 | 57.7 | 52 | 55.5 | |
| UnifiedReward-7Bparameters=7B2025.12 | 52 | 47.2 | 42.8 | 47.4 | |
| Qwen2.5-VL-7Bparameters=7B2025.12 | 51.8 | 45.4 | 41.1 | 46.1 | |
| Qwen3-VL-8Bparameters=8B2025.12 | 47.6 | 56.6 | 47.6 | 50.6 |