PhysBrain 1.5: From Vision-Language Models to Physical Foundation Models

Benchmark Model Rank Results
multimodal-reasoning-on-cv-benchPhysBrain 1.5-8B#1Accuracy: 90.0
spatial-reasoning-on-mmsi-benchPhysBrain 1.5-8B#2Accuracy: 41.0
spatial-reasoning-on-vsi-benchPhysBrain 1.5-8B#4Accuracy: 61.9
visual-question-answering-vqa-on-erqaPhysBrain 1.5-8B#4Accuracy: 52.8
visual-question-answering-vqa-on-mmePhysBrain 1.5-8B#8Score: 2330.07
zero-shot-video-question-answer-on-video-mme-w-o-subsPhysBrain 1.5-8B#30Accuracy (%): 68.22
vision-language-understanding-on-mmstarPhysBrain 1.5-8B#61Accuracy: 65.6