List your company

Visual Mathematical Reasoning on MathVista (Accuracy, Latency, Speedup)

75.4AccuracyDRL-trained M_d_dft (7B)
69.6871.16572.6574.135Jun 29, 2026

Evaluation results

MethodLinks
DRL-trained M_d_dft (7B)Draft (1 - pt)%=20%2026.0675.455.51.17
DRL-trained M_d_dft (7B)Draft (1 - pt)%=30%2026.0675.353.51.21
JRL-trained M_d_joint (7B)Draft (1 - pt)%=30%2026.0674.447.61.36
DRL-trained M_d_dft (7B)Draft (1 - pt)%=50%2026.0674.343.61.49
JRL-trained M_d_joint (7B)Draft (1 - pt)%=60%2026.0674.226.92.41
OursDraft (1 - pt)%=60%2026.0674.226.92.41
Mt(Skywork-R1V-38B)2026.0674.164.8—
JRL-trained M_d_joint (7B)Draft (1 - pt)%=40%2026.0674.140.11.62
JRL-trained M_d_joint (7B)Draft (1 - pt)%=50%2026.067432.61.99
Token-probability-based (Paradigm 1)Draft (1 - pt)%=60%2026.0672.229.52.19
Special-token-based (Paradigm 2)Draft (1 - pt)%=60%2026.0672.126.92.41
RandomDraft (1 - pt)%=60%2026.0671.627.52.36
JRL-trained M_d_joint (7B)Draft (1 - pt)%=100%2026.06704.9—
DRL-trained M_d_dft (7B)Draft (1 - pt)%=100%2026.0669.910.2—