List your company

Visually Grounded Reasoning on V* Bench (test)

95Overall Accuracyo3
61.40870.12978.8587.571Mar 4, 2026

Evaluation results

MethodLinks
o32026.0395——
DeepScan-72BBackbone=Qwen2.5-VL-72B, k=∞2026.0394.294.893.4
DeepScan-72BBackbone=Qwen2.5-VL-72B, k=102026.0393.793.993.4
DeepScanBackbone=Qwen2.5-VL-7B, k=∞2026.0391.193.986.8
DeepScanBackbone=Qwen2.5-VL-7B, k=102026.0390.69386.8
DeepEyes2026.039092.186.8
ViGoRL2026.0386.4——
TreeVGRevaluation_protocol=official codes, suffix=†2026.0385.986.185.5
Qwen2.5-VL-72BBackbone=Qwen2.5-VL-72B2026.0384.890.880.9
Dyfoevaluation_protocol=official codes, suffix=†2026.0384.382.686.8
Thyme-VL2026.0382.283.580.3
ZoomRefineevaluation_protocol=official codes, suffix=†2026.0382.285.377.6
PixelReasoner2026.0380.683.576.3
Dyfo-Qvariant=Q2026.038082.981.2
Seal2026.0374.876.375.4
Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.0374.377.469.7
Dyfo-Lvariant=L2026.0362.753.959.2