Multimodal Evaluation on MM (test)
51.6Rubric RewardPOW3R dynamic
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| POW3R dynamicBase policy=Qwen3-VL-8B2026.05 | 51.6 | 22.6 | |
| Category-balancedBase policy=Qwen3-VL-8B2026.05 | 50.4 | 21.6 | |
| Static scalarBase policy=Qwen3-VL-8B2026.05 | 49.5 | 20.7 | |
| POW3R dynamicBase policy=Qwen3-VL-4B2026.05 | 48.8 | 20.2 | |
| Category-balancedBase policy=Qwen3-VL-4B2026.05 | 47.9 | 18.7 | |
| Static scalarBase policy=Qwen3-VL-4B2026.05 | 47.1 | 17.9 | |
| BinaryBase policy=Qwen3-VL-8B2026.05 | 47 | 18.3 | |
| BaseBase policy=Qwen3-VL-8B2026.05 | 46.8 | 18.4 | |
| BinaryBase policy=Qwen3-VL-4B2026.05 | 42.8 | 14.6 | |
| BaseBase policy=Qwen3-VL-4B2026.05 | 42.6 | 14.7 | |
| POW3R dynamicBase policy=Gemma3-4B2026.05 | 37.8 | 10.3 | |
| Category-balancedBase policy=Gemma3-4B2026.05 | 36.7 | 9.4 | |
| Static scalarBase policy=Gemma3-4B2026.05 | 35.8 | 8.5 | |
| BaseBase policy=Gemma3-4B2026.05 | 34.1 | 7.7 | |
| BinaryBase policy=Gemma3-4B2026.05 | 33.9 | 7.6 |