List your company

Real-world Spatial Reasoning

40AccuracyGemini-2.5-Pro
24.428.4532.536.55Jun 11, 2026

Evaluation results

MethodLinks
Gemini-2.5-ProModel Type=Commercial2026.0640
GPT-5Model Type=Commercial2026.0638.3
Qwen3-32BModel Type=Open-Source2026.0638
Gemini-2.5-FlashModel Type=Commercial2026.0636.7
Qwen3-235BModel Type=Open-Source2026.0634.5
Qwen3-30BModel Type=Open-Source2026.0632.8
VTool-R1-7BModel Type=Tool-Augmented2026.0630.5
w/o RLModel Type=Ablation2026.0629.6
PERIA-8BModel Type=Proposed2026.0629.1
Mini-o3Model Type=Tool-Augmented2026.0628.9
InternVL3.5-8BModel Type=Open-Source2026.0628.4
w/o ToolModel Type=Ablation2026.0628.3
R1-OneVision-7BModel Type=Tool-Augmented2026.0628.2
Qwen3-8BModel Type=Backbone Baseline2026.0627.7
InternVL3.5-14BModel Type=Open-Source2026.0626.8
w/o SFTModel Type=Ablation2026.0625