MiniCPM-V 4.7 goes open-source: LES technique runs a 35B-A3B MLLM on an iPad at 17 tok/s
SimplyAnnisa · x · 2026-10-13
OpenBMB released MiniCPM-V 4.7, an open-source MLLM series pushing the performance-efficiency Pareto frontier for on-device multimodal AI:
- MiniCPM-V 4.7 35B-A3B: introduces LES (Lookahead Expert Scheduling), which predicts and preloads experts — billed as a milestone for on-device multimodal AI
- MiniCPM-V 4.7 1B: lightweight, easy to deploy
- Competitive results across 34 benchmarks covering multimodal understanding, STEM reasoning, OCR, and long-video comprehension
A user reports 17 tokens/sec on an iPad, joking her own output is "0.3 useful thoughts per hour."
Related event: OpenBMB Open-Sources MiniCPM-V 4.7: 35B MoE Multimodal on Device(7 posts)→
More from Infra
- oQe Quantization Update Cuts Code KL Divergence by Up to 18% Across 69 Models — huggingface · 2026-10-13
- 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