List your company
ResearchFormerly WizWand

Visual Probing on VisualProbe Hard

44.3AccuracyPERIA-8B
14.03621.89329.7537.607Jun 11, 2026
Updated 4mo ago

Evaluation results

MethodLinks
PERIA-8BModel Type=Proposed2026.0644.3
Mini-o3Model Type=Tool-Augmented2026.0641.5
Gemini-2.5-FlashModel Type=Commercial2026.0633.6
Gemini-2.5-ProModel Type=Commercial2026.0632
w/o RLModel Type=Ablation2026.0632
Qwen3-235BModel Type=Open-Source2026.0627.3
Qwen3-8BModel Type=Backbone Baseline2026.0626.6
w/o SFTModel Type=Ablation2026.0626.4
GPT-5Model Type=Commercial2026.0625.4
Qwen3-30BModel Type=Open-Source2026.0625.4
Qwen3-32BModel Type=Open-Source2026.0624.5
VTool-R1-7BModel Type=Tool-Augmented2026.0622.6
InternVL3.5-14BModel Type=Open-Source2026.0619.8
w/o ToolModel Type=Ablation2026.0619.8
R1-OneVision-7BModel Type=Tool-Augmented2026.0617.3
InternVL3.5-8BModel Type=Open-Source2026.0615.2