Visual Reasoning on VisualProbe Hard
0.434AccuracyDeepconf
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DeepconfBackbone=Qwen3-VL Instruct2026.02 | 0.434 | 14.7 | |
| SCBackbone=Qwen3-VL Instruct2026.02 | 0.425 | — | |
| ASCBackbone=Qwen3-VL Instruct2026.02 | 0.425 | 22.5 | |
| ESCBackbone=Qwen3-VL Instruct2026.02 | 0.425 | 6.7 | |
| RTWIBackbone=Qwen3-VL Instruct2026.02 | 0.425 | 16.2 | |
| CISCBackbone=Qwen3-VL Instruct2026.02 | 0.396 | 9.2 | |
| Self-Cer.Backbone=Qwen3-VL Instruct2026.02 | 0.377 | 12 | |
| DeepEyesBackbone=DeepEyes-7B2026.02 | 0.351 | — | |
| BaseBackbone=Qwen3-VL Instruct2026.02 | 0.349 | — | |
| SCBackbone=Qwen3-VL Thinking2026.02 | 0.302 | — | |
| ASCBackbone=Qwen3-VL Thinking2026.02 | 0.302 | 29.7 | |
| ESCBackbone=Qwen3-VL Thinking2026.02 | 0.302 | 8.6 | |
| RTWIBackbone=Qwen3-VL Thinking2026.02 | 0.302 | 44.6 | |
| DeepconfBackbone=Qwen3-VL Thinking2026.02 | 0.292 | 40.3 | |
| BaseBackbone=Qwen3-VL Thinking2026.02 | 0.283 | — | |
| CISCBackbone=Qwen3-VL Thinking2026.02 | 0.274 | 15 | |
| Self-Cer.Backbone=Qwen3-VL Thinking2026.02 | 0.274 | 14.5 | |
| GPT-4oBackbone=GPT-4o2026.02 | 0.112 | — |