List your company

Mathematical Visual Question Answering on MathVerse Vision-Only (testmini)

91Top-1 AccuracyDraw2Think CT
42.22454.88767.5580.213May 20, 2026

Evaluation results

MethodLinks
Draw2Think CTBackbone=gemini-3-flash-preview, Execution Environment=executable environment2026.0591
Draw2Think BLBackbone=gemini-3-flash-preview, Execution Environment=direct VLM2026.0587.5
Human expertNote=10 college students2026.0566.7
AVARBackbone=qwen2.5-vl-7b, Protocol=Fine-tuned, Mechanism=RL2026.0550.4
NeSyGeoBackbone=qwen2.5-vl-7b, Protocol=Fine-tuned, Mechanism=SFT2026.0546.7
SocraticGeoBackbone=qwen2.5-vl-7b, Protocol=Fine-tuned, Mechanism=GRPO2026.0545.1
Qwen2.5-VL-7BProtocol=Zero-shot2026.0544.1