STEP3-VL-10B Technical Report

Benchmark Model Rank Results
multimodal-reasoning-on-visulogicStep3-VL-10B (PaCoRe)#1Accuracy: 32.7
visual-question-answering-vqa-on-mmbench-cnStep3-VL-10B (PaCoRe)#1Accuracy: 91.96
visual-question-answering-vqa-on-mmbench-enStep3-VL-10B (PaCoRe)#1Accuracy: 92.38
mathematical-reasoning-on-hmmt-feb-2025Step3-VL-10B (PaCoRe)#4Accuracy: 92.14
multimodal-reasoning-on-mmmuStep3-VL-10B (PaCoRe)#11Accuracy: 80.11
visual-question-answering-vqa-on-blinkStep3-VL-10B (PaCoRe)#11Accuracy: 67.39
multimodal-reasoning-on-mathvisionStep3-VL-10B (PaCoRe)#13Accuracy: 75.95
multimodal-reasoning-on-mathvistaStep3-VL-10B (PaCoRe)#14Accuracy: 85.5
vision-language-understanding-on-mmstarStep3-VL-10B (PaCoRe)#15Accuracy: 77.64
multiple-choice-question-answering-mcqa-on-supergpqaStep3-VL-10B (PaCoRe)#16Accuracy: 53.15
chart-question-answering-on-charxiv-rqStep3-VL-10B#17Accuracy: 59.52
object-counting-on-countbenchStep3-VL-10B (PaCoRe)#18Accuracy: 88.8
visual-question-answering-vqa-on-realworldqaStep3-VL-10B (PaCoRe)#23Accuracy: 75.56
computer-use-agents-on-screenspot-v2Step3-VL-10B#24Accuracy: 92.61
multimodal-reasoning-on-mmmu-proStep3-VL-10B (PaCoRe)#25Accuracy: 67.18
computer-use-agents-on-screenspot-proStep3-VL-10B#31Accuracy: 51.55
multi-task-language-understanding-on-mmlu-proStep3-VL-10B (PaCoRe)#53Accuracy: 77.09
multiple-choice-question-answering-mcqa-on-gpqa-diamondStep3-VL-10B (PaCoRe)#73Accuracy: 73.99