Step-GUI Technical Report

Benchmark Model Rank Results
computer-use-agents-on-osw-g-osworld-gStep-GUI-8B#3Accuracy: 70.0
visual-question-answering-on-v-benchStep-GUI-8B#5Accuracy: 89.0
chart-question-answering-on-charxiv-dqStep-GUI-8B#6Accuracy: 86.3
computer-use-agents-on-mmb-gui-mmbench-guiStep-GUI-8B#6Accuracy: 85.6
visual-question-answering-vqa-on-mmbench-cnStep-GUI-8B#6Accuracy: 88.0
visual-question-answering-vqa-on-mmbench-enStep-GUI-8B#6Accuracy: 89.1
computer-use-agents-on-screenspot-v2Step-GUI-8B#10Accuracy: 95.1
visual-question-answering-vqa-on-simplevqaStep-GUI-8B#10Accuracy: 50.8
computer-use-agents-on-screenspot-proStep-GUI-8B#11Accuracy: 62.6
optical-character-recognition-ocr-on-ocrbenchStep-GUI-8B#13Score: 880
chart-question-answering-on-charxiv-rqStep-GUI-8B#26Accuracy: 44.6
multimodal-reasoning-on-mathvistaStep-GUI-8B#48Accuracy: 74.4
vision-language-understanding-on-mmstarStep-GUI-8B#49Accuracy: 68.0