China Association for Science and Technology Unveils Two AI Models

What changed

China’s Association for Science and Technology unveiled two AI models on September 30 under its “New Tiangong Kaiwu” initiative:

• “Lizhi” Uni-MoE, a multimodal language model developed by a Harbin Institute of Technology (Shenzhen) team.
• “Caihua” MatVerse, a materials-science model developed by Suzhou National Laboratory with Shanghai AI Laboratory and Shanghai Jiao Tong University.

Both were presented as major technology achievements and added to a digital technology-achievement exhibition database. Lizhi also received the 2025 Wu Wenjun Artificial Intelligence Science and Technology Progress Award grand prize.

Why it matters

The announcement shows China presenting AI development as more than a race to build general-purpose chatbots. Lizhi targets language understanding, generation and cross-domain use. Caihua is designed for materials research, linking data, models, intelligent agents, evaluation and applications in one system.

That distinction matters. A model built for a research discipline can be judged by whether it helps scientists work through specialised problems, not simply by how fluent its answers sound. Caihua’s evaluation benchmark covers 19 materials-science subdisciplines, and the model is described as having 10 billion parameters.

The institutions involved say both models are already being used by leading Chinese universities or companies. But the announcement gives no independent performance benchmarks, deployment scale, revenue figures, access terms or contract details. The practical test will be whether these systems produce repeatable advances outside the launch stage, especially in the laboratories and companies now using them.