Visual Reasoning on Hull-Bench
67.3AccuracyUnsilencing Latent Reasoning
Evaluation Results
| Method | Links | |
|---|---|---|
| Unsilencing Latent ReasoningBackbone=Qwen2.5VL-7B2026.05 | 67.3 | |
| LVRBackbone=Qwen2.5VL-7B2026.05 | 66.67 | |
| DMLRBackbone=Qwen2.5VL-7B2026.05 | 65.83 | |
| MonetBackbone=Qwen2.5VL-7B2026.05 | 65.67 | |
| ICoTBackbone=Qwen2.5VL-7B2026.05 | 65.51 | |
| VanillaBackbone=Qwen2.5VL-7B2026.05 | 65.4 | |
| LVRRFBackbone=Qwen2.5VL-7B2026.05 | 65.19 | |
| CCoTBackbone=Qwen2.5VL-7B2026.05 | 64.88 | |
| ICoTBackbone=Qwen2.5VL-3B2026.05 | 64.67 | |
| DMLRBackbone=Qwen2.5VL-3B2026.05 | 64.67 | |
| CoVTBackbone=Qwen2.5VL-7B2026.05 | 64.46 | |
| Unsilencing Latent ReasoningBackbone=Qwen2.5VL-3B2026.05 | 64.35 | |
| VanillaBackbone=Qwen2.5VL-3B2026.05 | 64.14 | |
| CCoTBackbone=Qwen2.5VL-3B2026.05 | 64.03 | |
| MCoTBackbone=Qwen2.5VL-3B2026.05 | 63.8 | |
| MCoTBackbone=Qwen2.5VL-7B2026.05 | 63.62 | |
| LVRRFBackbone=Qwen2.5VL-3B2026.05 | 60.99 | |
| LVRBackbone=Qwen2.5VL-3B2026.05 | 59.31 |