List your company

Visual Mathematical Reasoning on MathVerse (test)

68.1AccuracyQwen3-VL-8B-Instruct + DUPL
57.1860.01562.8565.685Oct 1, 2025

Evaluation results

MethodLinks
Qwen3-VL-8B-Instruct + DUPLModel Size=8B, Optimization=DUPL2025.1068.1
Qwen3-VL-4B-Instruct + DUPLModel Size=4B, Optimization=DUPL2025.1065.1
Qwen3-VL-8B-Instruct + GRPOModel Size=8B, Optimization=GRPO2025.1063.4
Qwen3-VL-4B-Instruct + GRPOModel Size=4B, Optimization=GRPO2025.1062
Qwen3-VL-8B-InstructModel Size=8B2025.1058.6
Qwen3-VL-4B-InstructModel Size=4B2025.1057.6