Reward Modeling on ExeVR-Bench Android
83.5AccuracyExeVRM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ExeVRMParameters=8B2026.03 | 83.5 | 77.2 | 95 | |
| Seed-2.0 ProModel Category=Proprietary2026.03 | 78 | 82.6 | 71 | |
| GPT-5.2Model Category=Proprietary2026.03 | 74.5 | 75.5 | 75.8 | |
| Gemini 2.5 ProModel Category=Proprietary2026.03 | 74 | 81.6 | 62 | |
| Gemini 3 ProModel Category=Proprietary2026.03 | 73.5 | 74.7 | 71 | |
| ExeVRMParameters=4B2026.03 | 72.5 | 66.4 | 91 | |
| Qwen3-VLParameters=4B2026.03 | 68.5 | 70.8 | 63 | |
| Qwen3-VLParameters=8B2026.03 | 64.5 | 71.6 | 48 | |
| Qwen2.5-VLParameters=7B2026.03 | 63.5 | 62.7 | 66 | |
| InternVL-3.5Parameters=8B2026.03 | 57.5 | 59.3 | 48 | |
| LLaVA-Next-VideoParameters=7B2026.03 | 49 | 47.5 | 19 |