China AI Intelligence: Our views on Kimi K3 release and WAIC 2026

Research evidence excerpt

China AI Intelligence: Our views on Kimi K3 release and WAIC 2026

Global Research

20 July 2026ab

China AI Intelligence Equities

ChinaOur views on Kimi K3 release and WAIC 2026

Internet Services

Wei Xiong

Rapid progress in China AI model iteration; WAIC points to broader adoption Analyst

We are encouraged by the rapid progress of China models following the success of wei.xiong@ubs.com

Zhipu's GLM-5.2; the release of Moonshot's AI model, Kimi K3, has also attracted +86-21-3866 8883

strong global attention as the world’s largest open-weight LLM (Large Language Kenneth Fong

Model), with performance approaching leading closed-source frontier models. Alibaba Analyst

announced the Qwen 3.8-Max preview, which reportedly targets performance close kenneth-kc.fong@ubs.com

to Fable 5, while the potential release of DeepSeek V4 GA (the official version) could +852-3712 3890

strengthen China’s cost-efficiency advantage in models. The World Artificial Charles Chen

Intelligence Conference (WAIC) 2026 showcased broader China AI development Analyst

across foundation models, agent applications, consumer / enterprise AI products and S1460524020001

domestic compute infrastructure, suggesting the industry is moving from catch-up charles-za.chen@ubs.com

mode in model capability to broader AI adoption (see more detail inside). +86-21-3866 8907

Kimi K3 scale-up brings China closer to the global frontier

We highlight three key points from the Kimi K3 launch: 1) it is the largest open-

weight LLM globally with 2.8trn parameters and a 1m-token context window, which

we view as evidence of Moonshot’s commitment to pushing the model intelligence

ceiling through pre-training scale-up, consistent with its prior K-series large parameter