China has unveiled DeepZang, the world's first large language model for the Tibetan language, in a move that could broaden the use of Tibetan in digital communications and public services.

The model was launched in Lhasa, capital of the Xizang autonomous region, by local developer Choknor. It supports interactions in Tibetan, as well as standard Chinese and English, and integrates functions including AI dialogue, real-time translation and speech-to-text transcription.

At the launch ceremony, the World Record Certification Agency awarded DeepZang the certification of "The World's First Tibetan Large Language Model", recognizing its global pioneering status.

According to the company's founder, Tenzin Norbu, the model is the result of more than four years of development. So far, it has accumulated nearly 70 million entries of standardized parallel corpora and more than 30,500 hours of voice data covering Tibetan's three major dialects — Utsang, Kham and Amdo — forming what the company said is China's most extensive and accurately annotated Tibetan speech database to date.

Beyond its technical milestone, the platform is already moving toward practical deployment. Several institutions, including the Xizang branches of China Mobile and PICC Property and Casualty Co, have signed cooperation agreements to explore applications in sectors such as government services, education, healthcare and finance.

Li Yalong, deputy general manager of PICC Property and Casualty's Xizang branch, said the model could help overcome long-standing language barriers in serving farmers and herders.

"It will support the development of Tibetan language intelligent customer service and policy interpretation tools," he said, particularly in agriculture-related insurance services.

Academic users have also highlighted its potential. Sonam Yontan, a doctoral student at Xizang University, said the model has already improved efficiency in research.

"The system's translation and search functions are useful and convenient," he said. "It allows us to process materials and find sources much faster." He added that the model represents an unprecedented step forward for Tibetan in the field of artificial intelligence.

However, early user feedback suggests the technology remains a work in progress.

"DeepZang's Tibetan capabilities were clearly strong," said Nie Chang, a 26-year-old Tibetan-language teacher for non-native speakers."But its response speed felt slower than some mainstream products such as ChatGPT or Doubao."

He said some users encounter paywalls after limited queries, which may affect accessibility.

In educational settings, Nie added, the model has yet to function effectively as a teaching aid for nonnative students.

"It feels more like a search engine at this stage," he said, adding that explanations of grammar are not always clear or tailored to these learners' needs.

Online reactions have been mixed, with some users praising the platform for promoting Tibetan language and culture, while others pointed to issues such as translation accuracy, limited capability in handling complex queries and pricing barriers. Some also expressed hope that similar models could be developed for other languages, including Mongolian and Uygur.

The company acknowledged that the model is still evolving and said it will continue improving performance as more data and user feedback are incorporated. It also noted that the current mobile app is just one application built on the underlying model, which could be extended to devices such as smart glasses and earphones, as well as sector-specific solutions in healthcare and education.

Contact the writers at guoyanqi@chinadaily.com.cn