China's open-source AI models power real economy

Chinese artificial intelligence (AI) startup Moonshot AI on Monday fully open-sourced its latest flagship large language model, Kimi K3, marking another milestone in China's rapidly expanding open-source AI ecosystem.

With 2.8 trillion parameters, Kimi K3 is currently the world's largest open-source AI model by parameter count.

The release includes the model's complete weights, a technical report and three underlying infrastructure technologies used to train the model, allowing developers worldwide to download, deploy and further develop it locally.

While China has previously open-sourced AI models, this is the first time a flagship model at the 2.8-trillion-parameter scale has been fully released to the global developer community.

China's open-source AI accelerates

China's pace of open-sourcing large AI models has accelerated significantly this year, with several leading AI companies making their flagship models publicly available.

Within just six months, multiple Chinese developers have opened access to their top-tier foundation models.

According to Hugging Face's State of Open Source on Hugging Face: Spring 2026 report, Chinese-developed open-source models account for 41% of global downloads.

The report also showed that the top six most-called large language models worldwide are all developed by Chinese teams. Over the past 12 months, Chinese models have held the global record for the largest open-source model for nine months, maintaining the fastest pace of iteration.

From trillion-parameter models last year to the emergence of 1.6-trillion- and now 2.8-trillion-parameter open-source models, China has repeatedly set new records for the scale of open-source AI models.

China's open-source AI ecosystem now spans models ranging from hundreds of millions to hundreds of billions of parameters. Smaller models can run offline on smartphones and industrial equipment, while larger models are capable of handling complex research and analytical tasks, providing options for both edge devices and large enterprises.

From open models to industrial applications

Unlike closed-source models, which allow users to access AI capabilities only through service providers, open-source models make their model weights publicly available, enabling developers to download, deploy and customize them for their own applications.

China's open-source AI models have expanded beyond general-purpose assistants into industry-specific applications.

Dedicated open-source models have been developed for sectors including automobiles, machinery manufacturing, electric power and healthcare, embedding industry expertise and professional knowledge directly into AI systems and helping bridge the gap between general-purpose models and real-world industrial applications.

According to official data, cumulative downloads of Chinese open-source AI models have exceeded 10 billion, the highest in the world. Industry statistics show that six out of every 10 large-model downloads globally now involve models developed in China.

China's open-source AI ecosystem has evolved from isolated technological breakthroughs into a more comprehensive and mature ecosystem, according to the report.

As the technology continues to mature and deployment barriers fall, AI is moving beyond research laboratories and becoming increasingly integrated into manufacturing, energy, services and other sectors of the real economy, driving industrial upgrading and helping reduce business costs.

The open-source approach has also lowered barriers to industrial AI adoption, enabling large language models to evolve from general conversational tools into productivity tools capable of executing tasks, supporting industrial operations and improving efficiency across manufacturing, energy, transportation and finance.

Policy support

Zhong Xinlong, an associate researcher at the Future Industries Research Center of the China Center for Information Industry Development, said that in the past, advanced AI models were largely controlled by a small number of closed-source platforms, leaving enterprises to purchase AI capabilities.

As open-source models mature, companies can build customized AI applications by combining their own data, business processes and industry knowledge, creating solutions better suited to specific scenarios, he said.

China is also strengthening policy support for open-source AI.

Earlier this year, China's Ministry of Industry and Information Technology and seven other government departments jointly issued an implementation guideline for the "AI Plus Manufacturing" initiative.

The document identifies the development of high-level open-source AI communities and the deployment of benchmark open-source projects as key priorities. By 2027, it aims to promote the deep application of three to five general-purpose large models in manufacturing, build 100 high-quality industrial datasets and expand AI deployment across 500 representative industrial scenarios.

Zhong said open-source AI models are shifting global AI competition from dominance by a handful of technology giants toward broader industrial competition.

"AI sovereignty is becoming an important policy priority for many countries," he said. "Open-source development is expanding participation and enabling more languages, regions and industries to join the AI ecosystem."

China now holds 60% of the world's AI patents, the highest share globally. According to estimates by the Ministry of Industry and Information Technology, the country's core AI industry exceeded 1.2 trillion yuan in 2025, with more than 6,200 AI enterprises.

As China's open-source ecosystem continues to expand, AI is expected to become more deeply embedded across the real economy, providing sustained momentum for new industrialization and the high-quality development of the digital economy.