# multimodal models

As Cloud Giants Duel, China’s Mianbi Pushes a 9‑Billion‑Parameter Multimodal ‘Brain’ for Edge Devices

Mianbi Intelligence has launched MiniCPM‑o 4.5, a 9‑billion‑parameter multimodal model that can ingest continuous audio, video and text while producing simultaneous outputs, and introduced a Jetson‑based developer board called Pinea Pi. The company positions the stack as an early example of on‑device, embodied AI aimed at robotics, automotive and personal devices, arguing that hybrid cloud‑edge deployments will better meet latency, privacy and stability needs than cloud‑only approaches.