Your phone can now run a 35B large model! With only 3GB of memory, and lossless.

What happened: ModelBest released MiniCPM-V4.7, a multimodal on-device model series with 35B-A3B and 1B versions. The 35B-A3B version is described as the first open source model that runs smoothly on phones with no loss of capability. Key numbers: 35B total parameters, 3B activated per inference, about 3GB Why it matters: The company says the 35B-A3B version scores higher than Qwen3.6 35B-A3B in shorter inference time, and the 1B version beats Qwen3.5 0.8B on overall score and inference efficiency. Support covers Rockchip, Nvidia, Intel and Apple chips plus iOS, Android and HarmonyOS.

ModelBest released MiniCPM-V4.7, a multimodal on-device model series with 35B-A3B and 1B versions. The 35B-A3B version is described as the first open source model that runs smoothly on phones with no loss of capability.

The company says the 35B-A3B version scores higher than Qwen3.6 35B-A3B in shorter inference time, and the 1B version beats Qwen3.5 0.8B on overall score and inference efficiency. Support covers Rockchip, Nvidia, Intel and Apple chips plus iOS, Android and HarmonyOS.