# Openbmb releases MiniCPM-V 4.7-35B-A3B, first open-source MoE model for on-device use

Digest AI · Generative AI & Models · published 2026-10-12T17:16:21Z · updated 2026-10-12T18:03:02Z

Canonical: https://digestai.news/story/openbmb-releases-minicpm-v-4-7-35b-a3b-first-open-source-moe-model-for

## Summary

openbmb launched **MiniCPM-V 4.7-35B-A3B**, the first open-source mixture-of-experts (MoE) model optimized for **on-device inference** on phones and tablets. The model runs smoothly on an **iPhone 17 Pro Max** with **13+ tokens per second** and **3GB of memory**, thanks to techniques like **Lookahead Expert Scheduling (LES)**, fine-grained quantization, and **DSpark speculative sampling**. It supports **single-image, multi-image, and video understanding** while maintaining strong performance metrics, outperforming **Qwen3.6 35B-A3B** in **Instruct mode** with faster inference times and higher scores.

The model supports **iOS, Android, and HarmonyOS**, with **day-0 compatibility** for chips from **Rockchip, Houmo, Intel, AMD, and Arm**. It also integrates with frameworks like **SGLang, vLLM, llama.cpp, and Ollama**, offering **GGUF, BNB, AWQ, and GPTQ quantization variants**, including **FP8 support**. openbmb provides **Self-Replay Fine-Tuning (SRFT)** for low-cost customization. The model has seen **630 downloads last month** on Hugging Face.

## Key points

- First open-source MoE model for on-device use, running at 13+ tokens/s on iPhone 17 Pro Max with 3GB memory
- Outperforms Qwen3.6 35B-A3B in Instruct mode with faster inference and higher scores, per openbmb’s benchmarks
- Supports iOS, Android, HarmonyOS, and chips from Rockchip, Houmo, Intel, AMD, and Arm with day-0 compatibility

## Why it matters

This model bridges the gap between powerful AI and resource-constrained devices, enabling real-time multimodal tasks on phones and tablets without sacrificing performance. It could accelerate edge AI adoption for developers targeting mobile and embedded systems.

## Sources

1. [openbmb/MiniCPM-V-4.7-35B-A3B](https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B) (huggingface.co, 2026-10-12, primary source)
2. [openbmb/MiniCPM-V-4.7-1B](https://huggingface.co/openbmb/MiniCPM-V-4.7-1B) (huggingface.co, 2026-10-12, primary source)

## Cite

Digest AI, "Openbmb releases MiniCPM-V 4.7-35B-A3B, first open-source MoE model for on-device use", 12 October 2026, https://digestai.news/story/openbmb-releases-minicpm-v-4-7-35b-a3b-first-open-source-moe-model-for

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/openbmb-releases-minicpm-v-4-7-35b-a3b-first-open-source-moe-model-for.json