OpenBMB releases MiniCPM-V 4.7: a 35B MoE multimodal model running 13 tok/s on iPhone
AdinaYakup · x · 2026-10-13
OpenBMB has released the MiniCPM-V 4.7 family for on-device multimodal AI:
- 35B-A3B (on-device MoE): 35B total / 3B active parameters, runs at 13+ tokens/s on iPhone 17 Pro Max with 3GB RAM; OpenBMB claims it beats Qwen3.6 35B-A3B on both overall score and inference time.
- 1B ultra-lightweight VLM: outperforms Qwen3.5 0.8B on both performance and efficiency.
- Both support image/multi-image/video understanding, 4×/16× visual token compression, iOS/Android/HarmonyOS, and are Apache 2.0 licensed.
The models are live on Hugging Face with GGUF, FP8, GPTQ, AWQ and int4 variants.
Related event: OpenBMB Open-Sources MiniCPM-V 4.7: 35B MoE Multimodal On-Device(13 posts)→
More from Models
- LightOnAI's LightOnOCR-3 Demo Trends on Hugging Face Spaces — lightonai · 2026-10-13
- OpenAI Codex hit capacity errors all day for personal and corporate accounts — ssh4net · 2026-10-13
- OpenAI Publishes Math Progress Report and openai/math on GitHub — Two Minute Papers · 2026-10-13
- Gemini 4 Argon pricing ($2/$10) spotted on Google's SKU page, global release imminent — gaganghotra_ · 2026-10-13
- A style guide for making Claude-like models write normally, and its linguistic traps — gleech · 2026-10-13
- OpenAI denies Pro 200 grandfathering to a subscriber billed through the eligibility window — IrrelevantNecessity · 2026-10-13