Kimi K3 draws global attention to China's open-source AI

Chinese startup Moonshot AI on Thursday released Kimi K3, a new open-weight large language model featuring 2.8 trillion parameters, making it the world's largest model of its kind by parameter count, the company said.

Kimi K3 natively supports visual understanding and offers a context window of one million tokens. It has been designed for demanding applications including software engineering, knowledge-intensive work, deep research and multimodal understanding.

The company said the larger parameter count provides a higher capability ceiling for tackling complex tasks, while its internal evaluations indicate an overall performance approaching that of leading proprietary AI models.

Industry benchmarks place the model just behind Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol. Kimi K3 also topped a widely followed AI coding benchmark within hours of its release, according to the company.

Its debut is seen as another sign that Chinese AI developers are narrowing the gap with global industry leaders. The release comes amid growing interest in open-weight foundation models, which allow developers and enterprises to customize and deploy advanced AI systems more flexibly.

China opens up AI to all

Industry experts said Kimi K3 reflects a pivot in China's AI development toward open innovation. Rather than limiting access to proprietary systems, an increasing number of Chinese developers are making advanced models openly available, enabling researchers and businesses around the world to build applications on shared foundations. They said open collaboration helps accelerate technological progress while supporting broader access to AI capabilities.

Some observers noted that increasingly capable open-weight models could intensify competition between open and proprietary ecosystems, while others cautioned that performance claims based on company benchmarks will require broader third-party validation as developers begin testing the model in real-world applications.

The release also sparked debate in financial markets over the cost of developing frontier AI models and the scale of infrastructure investment, amid a broader sell-off in AI-related chip stocks.

The launch came as China stepped up its push for an AI commons. On Friday, Chinese authorities released an action plan on international AI cooperation together with a collection of case studies, covering practical collaboration, wider application of AI technologies and the principle of AI for good.

Industry experts said a growing number of Chinese open-source large language models are moving from individual breakthroughs toward collective advances, pointing the way for equitable and inclusive global AI development.