CodePercept: Code-Grounded Visual STEM Perception for MLLMs

Benchmark Model Rank Results
mathematical-reasoning-on-mathverseCodePercept (32B, S1)#3Accuracy: 73.56
multimodal-reasoning-on-mathvisionCodePercept (32B, S1)#18Accuracy: 69.96
multimodal-reasoning-on-mathvistaCodePercept (32B, S1)#43Accuracy: 75.9