V-Rubrics: Visual Faithfulness via Rubric-Based Reinforcement Learning

Benchmark Model Rank Results
mathematical-reasoning-on-mathverseV-Rubrics (+ GRPO) (extra data)#16Accuracy: 52.16
chart-question-answering-on-charxiv-rqV-Rubrics (+ GRPO) (extra data)#19Accuracy: 57.0
multimodal-reasoning-on-mathvistaV-Rubrics (GRPO w/ rubrics) (extra data)#22Accuracy: 81.3
multimodal-reasoning-on-mathvisionV-Rubrics (GRPO w/ rubrics) (extra data)#24Accuracy: 58.88
multimodal-reasoning-on-mmmuV-Rubrics (GRPO w/ rubrics) (extra data)#32Accuracy: 70.56
multimodal-reasoning-on-mmmu-proV-Rubrics (GRPO w/ rubrics) (extra data)#33Accuracy: 58.15