On January 24th, the China Judicial Artificial Intelligence Conference (CJAI2026) convened in Shanghai, where the Internet Judicial Research Institute of Tsinghua University officially launched LegalOne-R1. This innovative open-source legal large language model (LLM) was developed leveraging Huawei Cloud Ascend AI cloud services. LegalOne-R1 is available in three parameter configurations: 1.7 billion (1.7B), 4 billion (4B), and 8 billion (8B), catering to diverse computational needs.

As a generative legal reasoning model, LegalOne-R1 has undergone extensive training on authentic Chinese judicial data. Its development incorporated a sophisticated "mid-stage training + post-training" dual-phase enhancement approach, enabling the model to acquire a comprehensive array of legal capabilities. This meticulous process ensures that LegalOne-R1 excels in real-world scenarios, delivering unparalleled stability, accuracy, and practicality.

The model's development was generously supported by the "Huawei AI Hundred Schools Initiative" and esteemed industry collaborators. Notably, the LegalOne-R1-8B variant demonstrated exceptional legal acumen on public evaluation datasets, outperforming existing open-source models in key tasks and setting a new benchmark in the field.

LegalOne-R1 has been fully optimized for training and inference on the Huawei Ascend computing platform, utilizing the MindSpore framework. On January 23rd, the model's parameters were made publicly available on an open-source platform, accompanied by a comprehensive technical report and an application guide to facilitate seamless integration and utilization.

Open-source access: https://huggingface.co/CSHaitao/LegalOne-1.7B.