OpenBMB Releases MiniCPM-V 4.7: On-Device MoE Multimodal With 5x iPad Speedup
victormustar · x · 2026-10-13
OpenBMB open-sourced MiniCPM-V 4.7 in two variants: a 35B-A3B MoE model with new LES (Lookahead Expert Scheduling) achieving 5x speedup and 17+ tokens/s on iPad Pro M4 16GB with identical outputs, and a lightweight 1B model. Claimed competitive results across 34 benchmarks spanning multimodal understanding, STEM reasoning, OCR and long-video, pushing a new performance-efficiency Pareto frontier for on-device AI. Apache 2.0 on GitHub.
Related event: OpenBMB Open-Sources MiniCPM-V 4.7: 35B MoE Multimodal on Device(7 posts)→
More from Infra
- Rodolfo Rosini on 2040: cheap compute everywhere, devices jumping orders of magnitude — MikePFrank · 2026-10-13
- AI inference chips are a $225bn market: Vaire on why startups avoid silicon and its Nvidia-adjacent plan — MikePFrank · 2026-10-13
- Vaire CEO Explains Reversible Computing Chips That Recycle Energy Like a Tide Mill — MikePFrank · 2026-10-13
- China's $165B AI buildout rests on $11B model revenue and 4% of Nvidia's chip output — demian_ai · 2026-10-13
- New llama.cpp release adds multimodal model support and Metal performance gains — vboykis · 2026-10-13
- Broadcom optical roadmap: TH6 hits 102.4T in 2026, next-gen OCI by 2028 — firstadopter · 2026-10-13