ModelBest Releases MiniCPM5-2B Model for Edge Devices

ModelBest and the OpenBMB community announced in a press release the release of MiniCPM5-2B, an open source language model for edge devices. ModelBest says Artificial Analysis ranked it first on the Intelligence Index among open source models with fewer than 4 billion parameters, while it scored 20 on the Agentic Index.

The 2 billion parameter model supports tool calling, deep search, code generation, and reasoning across multiple steps. It can run locally on PCs, smartphones, robots, and IoT hardware for tasks including document processing, data synthesis, coding, and conversations across multiple turns.

ModelBest and OpenBMB also released datasets, training recipes, reinforcement learning infrastructure, and resources covering data curation, pretraining, and alignment. Downloads across the MiniCPM model family have exceeded 50 million.