Reward Modeling on ManiRewardBench Single-arm YAM
0.945Mean VOCTOPReward
Evaluation results
| Method | Links | |
|---|---|---|
| TOPRewardBackbone=Qwen3-VL-8B2026.02 | 0.945 | |
| GVLBackbone=Gemini-2.5-Pro, Chat template enforced=true2026.02 | 0.752 | |
| TOPRewardBackbone=Molmo2-8B2026.02 | 0.642 | |
| GVLBackbone=Qwen3-VL-8B2026.02 | 0.544 | |
| TOPRewardBackbone=Gemini-2.5-Pro, Chat template enforced=true2026.02 | 0.488 | |
| GVLBackbone=Molmo2-8B2026.02 | -0.017 |