Reward Modeling on ExeVR-Bench Ubuntu Human
84AccuracyExeVRM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ExeVRMParameters=8B2026.03 | 84 | 84 | 84 | |
| ExeVRMParameters=4B2026.03 | 81 | 82.9 | 78 | |
| Seed-2.0 ProModel Category=Proprietary2026.03 | 77.2 | 81.7 | 69.1 | |
| GPT-5.2Model Category=Proprietary2026.03 | 74 | 84.3 | 59 | |
| Qwen3-VLParameters=4B2026.03 | 72.5 | 76.5 | 65 | |
| Gemini 3 ProModel Category=Proprietary2026.03 | 71.2 | 71.7 | 71 | |
| Qwen3-VLParameters=8B2026.03 | 69 | 86.5 | 45 | |
| Gemini 2.5 ProModel Category=Proprietary2026.03 | 64.5 | 82.2 | 37 | |
| Qwen2.5-VLParameters=7B2026.03 | 63 | 65.1 | 56 | |
| InternVL-3.5Parameters=8B2026.03 | 52 | 52.2 | 48 | |
| LLaVA-Next-VideoParameters=7B2026.03 | 48.5 | 46.7 | 21 |