ModelBest launches MiniCPM-V 4.7, an open-source MoE model that runs on phones with just 3GB memory

ModelBest has released MiniCPM-V 4.7, a new on-device multimodal AI model in two sizes: 35B-A3B and 1B. The 35B-A3B is the first MoE model in the MiniCPM-V series.

Built on new LES inference acceleration tech, it is the first open-source MoE model to run smoothly on phones. It reaches over 13 tokens per second with just 3GB of memory on the iPhone 17 Pro Max, and works with iOS, Android, and HarmonyOS, plus Intel, AMD, Arm, and Rockchip chips.

In tests, the 35B-A3B version scored higher than Qwen3.6 35B-A3B with shorter inference time. The 1B version also beats Qwen3.5 0.8B in overall scores and speed.

The company also open-sourced its SRFT fine-tuning solution, helping developers expand model skills at low cost.