# ModelBest Open-Sources Edge Language Base Model MiniCPM5-2B

- Category: Flash (快讯)
- Published: 2026-09-08T11:54:00.000Z
- Source date: 2026-09-08
- HTML: https://www.88z.ai/en/news/flash-20260908-5e25460b7b7ecc27
- Markdown: https://www.88z.ai/en/news/flash-20260908-5e25460b7b7ecc27.md
- Other language: https://www.88z.ai/news/flash-20260908-5e25460b7b7ecc27.md

ModelBest and OpenBMB have released MiniCPM5-2B, an edge base model that ranks first among open-source models under 4B parameters on the AA leaderboard. It achieved a comprehensive score of 23 and an Agentic Index of 20. The project also open-sourced UltraData, Meshy framework, and JustRL II, with Day0 adaptation for Intel, Rockchip, and Arm.

## At a glance

- MiniCPM5-2B ranks first among open-source models under 4B on the AA leaderboard, with an Agentic Index about ten times higher than peers.
- The project open-sourced UltraData, Meshy, and JustRL II, completing Day0 adaptation for Intel, Rockchip, and Arm.

## The update

ModelBest and OpenBMB have jointly open-sourced MiniCPM5-2B, a 2-billion parameter base model designed for edge applications. The model performs exceptionally well on the AA leaderboard, ranking first among global open-source models with fewer than 4 billion parameters, achieving a comprehensive score of 23. Notably, its Agentic Index reached 20, representing performance approximately ten times that of similar models. In addition to the model itself, the project has simultaneously released the UltraData dataset, the Meshy framework, and JustRL II technology. Regarding hardware ecosystem support, MiniCPM5-2B has achieved Day0-level adaptation for Intel, Rockchip, and Arm platforms, ensuring developers can quickly deploy and utilize the model across various hardware devices.

## Sources

- [面壁智能](https://mp.weixin.qq.com/s/QjIQhrayJK47SAN2E5TpdA) (2026-09-08)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.