OpenBMB's new 1B vision language model pushes the performance-efficiency frontier for on-device multimodal AI, beating Qwen3.5 0.8B while running on phones.

Full article content could not be extracted automatically. Read the original below.