Reward Modeling on ExeVR-Bench Ubuntu Agent
85.1AccuracySeed-2.0 Pro
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Seed-2.0 ProModel Category=Proprietary2026.03 | 85.1 | 85.1 | 85.1 | |
| Gemini 2.5 ProModel Category=Proprietary2026.03 | 84.6 | 85.6 | 82.8 | |
| GPT-5.2Model Category=Proprietary2026.03 | 82.5 | 85.1 | 78.7 | |
| ExeVRMParameters=8B2026.03 | 82.5 | 85.9 | 77.7 | |
| Gemini 3 ProModel Category=Proprietary2026.03 | 80.4 | 75 | 90 | |
| ExeVRMParameters=4B2026.03 | 77.8 | 83.3 | 69.1 | |
| Qwen3-VLParameters=4B2026.03 | 76.2 | 72.9 | 82.9 | |
| Qwen3-VLParameters=8B2026.03 | 71.4 | 74.4 | 64.9 | |
| InternVL-3.5Parameters=8B2026.03 | 64.6 | 64.8 | 55.9 | |
| Qwen2.5-VLParameters=7B2026.03 | 64.6 | 65.5 | 60.6 | |
| LLaVA-Next-VideoParameters=7B2026.03 | 50.8 | 51.6 | 17 |