MiniCPM-V 4.7 collection hits Hugging Face with 35B-A3B and 1B in six quantization formats
Xianbao_QIAN · x · 2026-10-13
OpenBMB's MiniCPM-V 4.7 collection is live on Hugging Face, billed as a new chapter in on-device multimodal AI. It centers on the 35B-A3B (image-text-to-text) and 1B models, each shipped in GGUF, FP8, GPTQ, AWQ, int4, and DSpark formats, drawing hundreds of likes within the first hour. The collection also includes MiniCPM5, VoxCPM, AgentCPM, and MiniCPM-Robot lines.
Related event: OpenBMB Open-Sources MiniCPM-V 4.7: 35B MoE Multimodal on Device(7 posts)→
More from Multimodal
- AI short film "Mythical Beast Kitchen": fluffy creatures gaze at a giant fish — bLod2t · 2026-10-13
- AI-generated video recreates the first 1903 Tour de France with striking vintage realism — Fuhrmannski · 2026-10-13
- Surreal AI-Generated 'Weird Buildings' Look Eerily Photorealistic — miilesus · 2026-10-13
- The Real Test for Muse Wonder: Everyday Tasks, Not Minecraft Demos — alexandr_wang · 2026-10-13
- Scale CEO Alexandr Wang Endorses Muse Wonder for Design Work — alexandr_wang · 2026-10-13
- MiniMax H3 turns a storyboard draft into a slick cyberpunk robot video — ExpressWarthog8505 · 2026-10-13