Perception-demanding tasks on BLINK
61AccuracyPhi4-multimodal + DRScaffold
Evaluation Results
| Method | Links | |
|---|---|---|
| Phi4-multimodal + DRScaffoldScale=5.6B2026.05 | 61 | |
| Phi4-multimodalScale=5.6B2026.05 | 60.6 | |
| Qwen2.5-VLScale=32B2026.05 | 58.1 | |
| Qwen2.5-VL + DRScaffoldScale=3B2026.05 | 48.8 | |
| Qwen2.5-VLScale=3B2026.05 | 47.9 | |
| InternVL2.5Scale=2B2026.05 | 44.2 | |
| InternVL2.5 + DRScaffoldScale=2B2026.05 | 42.4 |