List your company
ResearchFormerly WizWand

Microscopic Reasoning on MicroVQA

59.2AccuracyGemini-2.5-Pro
37.98443.4924954.508May 21, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Gemini-2.5-ProModel Type=Closed-Source2026.0559.2
Gemini-2.5-FlashModel Type=Closed-Source2026.0558.6
GPT-5Model Type=Closed-Source2026.0557
MAESTROModel Type=Ours2026.0553
Qwen3-VL-32BModel Type=Open-Source & Baselines2026.0552.6
Kimi-K2.5Model Type=Open-Source & Baselines2026.0551.6
GLM-4.6VModel Type=Open-Source & Baselines2026.0551
PixelReasonerModel Type=Think with Images Methods2026.0550.8
Visual-ARFTModel Type=Think with Images Methods2026.0550.4
VTS-VModel Type=Think with Images Methods2026.0549.4
VisionReasonerModel Type=Think with Images Methods2026.0549
DeepEyesModel Type=Think with Images Methods2026.0548.8
ThymeModel Type=Think with Images Methods2026.0548.8
GPT-4oModel Type=Closed-Source2026.0547.8
Chain-of-FocusModel Type=Think with Images Methods2026.0547.4
MathCoder-VLModel Type=Think with Images Methods2026.0544
VTOOL-R1Model Type=Think with Images Methods2026.0543.8
DeepEyes-v2Model Type=Think with Images Methods2026.0541.4
Direct AnsweringModel Type=Open-Source & Baselines2026.0540.8
Untrained ModelModel Type=Open-Source & Baselines2026.0538.8