BEIJING, July 29 (Xinhua) -- China is making significant strides in reshaping the global AI landscape, as a new generation of frontier open-weight models is poised to bridge the global "intelligence gap" by providing advanced and affordable computing tools for less developed nations and grassroots innovators.
Moonshot AI, a Beijing-based startup, on Monday unveiled the full suite of its flagship Kimi K3 system, including the model weights, technical report, and the key infrastructure that powered its training.
The landmark model features 2.8 trillion parameters, built on a Mixture-of-Experts (MoE) framework, and is equipped with native image comprehension capabilities and an unprecedented 1-million-token context window to process lengthy texts seamlessly.
It does not rely on parameter expansion alone. Through a self-developed computing architecture and training optimization, the team increased computing efficiency by 250 percent despite constrained hardware resources, generating far more intelligent output per unit of computational power, according to a statement on the company's official WeChat account.
From now on, developers across the globe are permitted to freely download, run and adapt the model locally for internal research or end-user product creation, the statement said.
Classified as the world's largest open-weight AI model to date, Kimi K3 ranks third in overall intelligence on benchmark tests conducted by the independent evaluator Artificial Analysis, trailing behind Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol, two leading closed-source models. Its application programming interface pricing is just one-third of Claude Fable 5's, while delivering comparable coding performance.
Hot on Moonshot AI's heels, Alibaba's Qwen team unveiled a preview edition of its next-generation large language model, Qwen3.8, and announced that it will soon fully launch and open-weight the 2.4-trillion-parameter system.
The development of Chinese large models has directly challenged the "closed" path championed by Western industry leaders. The latter's traditional approach keeps cutting-edge technology proprietary, creating an oligopoly in which a few companies and nations monopolize AI's benefits and risks, widening the global governance gap.
The proliferation of China's open-weight AI ecosystems has yielded tangible results worldwide. Data from Hugging Face, a New York-based open-source AI platform, showed that in the past year Chinese models quickly accounted for the plurality, or 41 percent of downloads.
Cumulative worldwide downloads of Chinese open-weight large language models have surpassed 10 billion, while domestic platforms such as AtomGit have attracted over 11 million registered users, according to the People's Daily on July 24.
Hugging Face data also showed that Alibaba's models, including its Qwen family, had the most user-generated variants, more than models from Google and Meta combined this March, according to an MIT Technology Review report.
Countries across Southeast Asia and beyond are embracing Chinese open-weight systems. Singapore's government-backed AI Singapore program chose Qwen to build its latest regional model. In 2025, Malaysia announced that its sovereign AI ecosystem would run on DeepSeek, another prominent Chinese large model.
China's push for open-source AI aligns with its broader vision for global AI governance. At the recently concluded World Artificial Intelligence Conference (WAIC) 2026, China proposed building a fair and reasonable global AI governance system.
A statement issued at the conference enshrined "open source" as a vital pathway to inclusive AI development, calling for responsible efforts to build an open ecosystem and foster an inclusive international community.
To support this vision, China announced several initiatives over the next five years, including providing 5,000 AI training slots for developing countries, establishing international AI application cooperation centers with the Association of Southeast Asian Nations, the League of Arab States, the African Union, the Community of Latin American and Caribbean States, the Shanghai Cooperation Organization and BRICS, as well as promoting the deployment of its "Mazu" intelligent weather warning system in 30 countries.
This collaborative spirit was formalized as 29 nations signed an agreement in Shanghai to establish the World AI Cooperation Organization, aimed at supporting Global South nations in keeping pace with the technological revolution and bridging the intelligence divide.
"The vitality of technology lies in openness and flow," said Liu Tieyan, president of Zhongguancun Academy. "A number of Chinese open-weight large language models have evolved from isolated debuts to collective breakthroughs, offering new solutions and pathways for the global development of AI." ■
