The debut of a new Tibetan language artificial intelligence model is expected to expand digital access and public services across Tibetan-speaking regions, offering enhanced support for sectors ranging from healthcare and education to governance and cultural preservation.
Unveiled in Beijing on Wednesday, the model — named Zeta — was developed by the State Key Laboratory of Tibetan Intelligence at Qinghai Normal University. It has been filed under national regulations for generative AI services and has entered public testing and pilot application, becoming Qinghai's first large-scale model designed for full multimodal processing of the Tibetan language.
Dorlha, the lab's executive deputy director, said the regulatory filing marks a new starting point for applying the technology to real-world needs. Unlike traditional AI systems typically designed for text translation or speech recognition, Zeta takes a comprehensive, multidimensional approach to language processing.
"Its advanced capabilities span listening, speaking, reading, writing and translating across the three main Tibetan dialects — Amdo, U-Tsang and Kham," Dorlha said.
The system incorporates innovations such as mixed-language document recognition, audiobook production, ancient literature retrieval and intelligent subtitle transcription, among others, he said.
In addition, industry-specific features like digital broadcasting and subtitle transcription make it a valuable tool for sectors ranging from media to agriculture and tourism.
Technologically, Zeta integrates Tibetan, standard Chinese and English within a multilingual framework. It is supported by an expansive dataset including a 150-gigabyte high-quality Tibetan text corpus, 87 million multilingual parallel sentences and 30,000 hours of multi-dialect audio recordings.
Coupled with self-developed algorithms and compatibility with domestic AI infrastructure, the model demonstrates technical maturity and scalability, Dorlha said.
With versions available in 7 billion, 50 billion and 122 billion parameters, Zeta is designed to tackle a range of challenges, including linguistic translation, voice recognition, medical assistance and literature preservation.
Nyima Tashi, a professor at Xizang University and director of the State Key Laboratory of Tibetan Intelligence, said the launch of Zeta and its application products will contribute to high-quality economic and social development in Tibetan regions of China.
"The laboratory will continue to deepen Tibetan language AI research, increase scientific investment, strengthen talent development and enhance collaboration among industry, academia and research institutions," he said.
Moving forward, the team aims to refine the model by deploying multimodal capabilities through open application programming interfaces and fostering collaboration across academic institutions and private enterprises, building a broader ecosystem for Tibetan-language AI.
Zeta's release comes about a month after the launch of Deep-Zang, the first Tibetan large language model launched in the Xizang autonomous region, offering more options for users.
Tenzin Palden, a Tibetan student at Shandong Agricultural University, said Zeta offers hope for preserving his linguistic identity in an increasingly digital world.
"By addressing historical challenges like limited datasets and diversity in Tibetan dialects, this innovation provides much-needed momentum for bridging the wisdom of Tibetan traditions with modern development," he said.
"It is not just a technological achievement but also a reflection of the protection and transmission of ethnic culture."