{"version":1,"type":"story","url":"https://digestai.news/story/openbmb-releases-minicpm-v-4-7-35b-a3b-first-open-source-moe-model-for","json":"https://digestai.news/story/openbmb-releases-minicpm-v-4-7-35b-a3b-first-open-source-moe-model-for.json","markdown":"https://digestai.news/story/openbmb-releases-minicpm-v-4-7-35b-a3b-first-open-source-moe-model-for.md","slug":"openbmb-releases-minicpm-v-4-7-35b-a3b-first-open-source-moe-model-for","headline":"Openbmb releases MiniCPM-V 4.7-35B-A3B, first open-source MoE model for on-device use","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.\n\nThe 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.","keyPoints":["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"],"whyItMatters":"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.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["openbmb"],"models":["MiniCPM-V 4.7-35B-A3B","Qwen3.6 35B-A3B","MiniCPM-V 4.7-35B-A3B-DSpark"],"people":[]},"firstPublishedAt":"2026-10-12T17:16:21Z","updatedAt":"2026-10-12T18:03:02Z","sourceCount":2,"hasPrimarySource":true,"sources":[{"outlet":"huggingface.co","title":"openbmb/MiniCPM-V-4.7-35B-A3B","url":"https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B","publishedAt":"2026-10-12T17:16:21Z","type":"primary","primary":true,"lead":true},{"outlet":"huggingface.co","title":"openbmb/MiniCPM-V-4.7-1B","url":"https://huggingface.co/openbmb/MiniCPM-V-4.7-1B","publishedAt":"2026-10-12T18:03:02Z","type":"primary","primary":true,"lead":false}],"sourceNotes":null,"discussions":[{"site":"Reddit","url":"https://www.reddit.com/r/LocalLLaMA/comments/1x492f9/openbmbminicpmv471b/","points":null},{"site":"Reddit","url":"https://www.reddit.com/r/LocalLLaMA/comments/1x47taj/openbmbminicpmv4735ba3b/","points":null}],"thread":null,"cite":{"text":"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","publisher":"Digest AI","title":"Openbmb releases MiniCPM-V 4.7-35B-A3B, first open-source MoE model for on-device use","datePublished":"2026-10-12T17:16:21Z","url":"https://digestai.news/story/openbmb-releases-minicpm-v-4-7-35b-a3b-first-open-source-moe-model-for"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"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"}