Capek 0.5: An Execution-Centric Vision-Language Model for Embodied Intelligence

Benchmark Model Rank Results
code-generation-on-livecodebench-v6Capek0.5-35B-A3B–Pass@1: 0.7291
multi-task-language-understanding-on-mmlu-proCapek0.5-35B-A3B–Accuracy: 84.77
video-question-answering-on-mvbenchCapek0.5-35B-A3B–Avg.: 71.32
visual-question-answering-vqa-on-realworldqaCapek0.5-35B-A3B–Accuracy: 83.4
zero-shot-video-question-answer-on-video-mme-w-o-subsCapek0.5-35B-A3B–Accuracy (%): 74.19