ModelBest Releases Open-Source MiniCPM5-2B for Edge Devices

BEIJING, Sept. 8, 2026 /CourierPR/ -- ModelBest, a leading AI startup based in China, has introduced a groundbreaking open-source language model named MiniCPM5-2B, designed to run on edge devices. The release, announced on September 9th in collaboration with the OpenBMB open-source community, aims to bring advanced AI capabilities to edge devices at a significantly reduced cost.

MiniCPM5-2B, despite its modest 2 billion parameters, exhibits remarkable versatility and functionality. The model supports tool calling, deep search, code generation, and multi-step reasoning, setting new standards for general-purpose agentic capabilities. In a recent evaluation by Artificial Analysis, MiniCPM5-2B secured the top spot on the Intelligence Index among open-source models with fewer than 4 billion parameters. It also scored 20 on the Agentic Index, outperforming its peers in the sub-4B category.

This development signifies a shift in AI development strategies, focusing on increasing "intelligence density" rather than simply scaling up the number of parameters. By doing so, ModelBest and OpenBMB are enabling resource-constrained devices to achieve powerful performance within minimal compute, memory, and power requirements. This approach is expected to accelerate the transition of advanced AI from centralized clouds to everyday edge hardware, providing significant benefits for both end-users and enterprises.

The practical applications of MiniCPM5-2B are diverse and include document processing, data synthesis, code generation, and complex multi-turn Q&A. These capabilities can now be run locally on a variety of devices, including PCs, smartphones, robotics, and IoT hardware. Shifting compute to the edge not only enhances user privacy and data sovereignty but also significantly reduces latency and cloud API costs, while decreasing infrastructure dependency.

ModelBest and OpenBMB have taken a step further by fully open-sourcing their end-to-end pipeline, from data curation to training and alignment. This radical transparency provides global researchers and developers with a replicable blueprint to fine-tune and deploy specialized edge applications. To date, downloads of the MiniCPM family have surpassed 50 million, underscoring the company's commitment to democratizing lightweight and deployable edge AI technology worldwide.

The release of MiniCPM5-2B marks a significant milestone in the evolution of edge AI, offering a more efficient and accessible solution for a wide range of applications. As ModelBest continues to push the boundaries of what can be achieved on edge devices, the landscape of AI deployment is set to undergo further transformation.