Face intelligence launches new end-device multimodal model minicpm-v 4.7 with enhanced performance

On October 13, 2026, Face Intelligence announced the release of its latest end-device multimodal model, MiniCPM-V 4.7, offering two variants: 35B-A3B and 1B. The 35B-A3B version is the first MoE model in the MiniCPM-V series to support seamless, lossless operation on end devices, achieving a decoding speed of 13+ token/s with just 3GB of memory on the iPhone 17 Pro Max. The model supports iOS, Android, and HarmonyOS, and is compatible with Intel, AMD, Arm, Rockchip, and Huami chipsets, covering a wide range of mobile and local computing scenarios. It also includes an open-source Self-Replay Fine-Tuning (SRFT) framework to aid developers in expanding model capabilities and reducing forgetting. The 35B-A3B version outperforms Qwen3.6 35B-A3B in Instruct mode, while the 1B variant shows significant improvements over MiniCPM-V 4.6, surpassing Qwen3.5 0.8B in both performance and efficiency. The model's documentation is available on GitHub at MiniCPM-V/README_zh.md at main · OpenBMB/MiniCPM-V.