ModelBest has released MiniCPM-V 4.7, a new generation of its on-device multimodal model family, in two sizes: 35B-A3B and 1B. The 35B-A3B version is the first mixture-of-experts model in the MiniCPM-V series. The weights, code and documentation are published through the OpenBMB GitHub repository under the Apache-2.0 license.
The headline claim is efficiency on phones. ModelBest says the 35B-A3B model, which activates about 3 billion parameters per token, runs on an iPhone 17 Pro Max using only about 3GB of memory while decoding at more than 13 tokens per second. The company attributes this to a new inference acceleration technique it calls LES, or lookahead expert scheduling, and describes the model as the first open-source MoE model that can run losslessly and smoothly on device.
On benchmarks reported by the team, the 35B-A3B model scores 91.72 on MMBench, 91.2 on OCRBench and 90.21 on CMMLU. In Instruct mode, ModelBest says it reaches a higher overall score than Qwen3.6 35B-A3B with shorter reasoning time. The 1B version improves on MiniCPM-V 4.6 in basic capabilities, and the company says it leads Qwen3.5 0.8B on both overall score and inference efficiency.
In hands-on tests published by QbitAI, the model handled chart reasoning and converted academic papers into Markdown with the phone offline, a use case aimed at private, local document work.
MiniCPM-V 4.7 supports iOS, Android and HarmonyOS, and has been adapted for chip platforms from Intel, AMD, Arm, Rockchip and Houmo AI, covering both mobile devices and local PCs. Alongside the model, ModelBest open-sourced Self-Replay Fine-Tuning, or SRFT, a fine-tuning recipe meant to help developers add new capabilities at low cost while limiting forgetting of existing skills.
The release continues a steady cadence for the MiniCPM line, following MiniCPM-V 4.6 in May and the MiniCPM5-2B text model in September.
SEE ALSO: MiniCPM-V 4.6: Tsinghua Spinoff Open-Sources a 1.3B Multimodal Model That Runs on a Single RTX 4090 · OpenBMB Releases MiniCPM5-2B as Open On-Device SOTA Under 4B