Reward Modeling on ExeVR-Bench Mac/Win
90AccuracyExeVRM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ExeVRMParameters=4B2026.03 | 90 | 87.7 | 93 | |
| ExeVRMParameters=8B2026.03 | 89 | 85.5 | 94 | |
| Seed-2.0 ProModel Category=Proprietary2026.03 | 81 | 86.1 | 74 | |
| Gemini 3 ProModel Category=Proprietary2026.03 | 75.6 | 75.3 | 76.8 | |
| GPT-5.2Model Category=Proprietary2026.03 | 74.5 | 88.9 | 68.7 | |
| Qwen3-VLParameters=4B2026.03 | 74.5 | 74.3 | 75 | |
| Gemini 2.5 ProModel Category=Proprietary2026.03 | 69 | 83.9 | 47 | |
| Qwen2.5-VLParameters=7B2026.03 | 68 | 74.3 | 55 | |
| Qwen3-VLParameters=8B2026.03 | 66.5 | 81.1 | 43 | |
| InternVL-3.5Parameters=8B2026.03 | 59.5 | 58.6 | 65 | |
| LLaVA-Next-VideoParameters=7B2026.03 | 49.5 | 48.8 | 21 |