Photo: Courtesy of Moonshot AI

After Kimi K3, a popular open-source AI model created by China's upstart Moonshot AI Co with 2.8 trillion parameters, has received massive visits than expected since its release last week, the company said on Sunday that it is now facing mounting computing-power challenges, and it has decided to suspend new C-end subscriptions from Sunday.

Chinese analysts said that from DeepSeek V4 to the newly released Kimi K3, numerous Chinese large language models (LLMs) have achieved breakthrough innovations across multiple performance metrics, earning strong user adoption. They also noted that China continues to emphasize an open-source and inclusive approach to AI development, aiming to make artificial intelligence a technology that truly benefits all of humanity.

According to Moonshot AI, over the past 48 hours, user visit request volume has significantly exceeded its projections, which is approaching the capacity limit. "To ensure a stable experience for our current subscribed users, we have decided to immediately suspend new C-end subscriptions," the company announced.

"All available computing resources will be dedicated to serving existing subscribers, ensuring their full rights and experiences remain unaffected," the company said, adding that it is trying to accelerate computing-power capacity expansion "at full speed". 

As new capacity comes online, Moonshot will gradually open more subscription slots until normal subscription services are fully restored, the company noted.

According to a post the company published on its WeChat account on Friday, Kimi K3 is the largest open-source model in the world by parameter scale, which natively supports visual understanding and can process up to 1 million tokens of context.

Moonshot said that Kimi K3 is designed for demanding tasks including software engineering, knowledge work, deep research and multimodal understanding.

In the Frontend Code Arena global AI leaderboard, Kimi K3 topped the ranking with 1679 score, surpassing Claude Fable 5 to take first place. 

"This may be the single biggest release of the year," and marks a moment when open-source Chinese AI models are surpassing closed US models, said Anastasios Angelopoulos, co-founder and CEO of Arena, a platform for evaluating AI systems, according to a report by AP News.

Kimi K3 topped the charts in Arena's ranking of what it calls "front-end coding capability," a metric of an AI large language model's performance. "More results are rolling in that are likely to continue to show it is at the top of the pack," Angelopoulos said.

Hu Qimu, a professor at the Maritime Silk Road Institute of Huaqiao University, told the Global Times on Monday that leveraging China's vast and rich data resources for iteration and refinement, Chinese AI models have achieved independent innovation breakthroughs across multiple performance metrics, earning strong user preference for their high cost-performance ratio.

Alibaba announced on Sunday on its WeChat account that its Qwen 3.8 is set to be released and open-sourced soon. The new model features 2.4 trillion parameters and is evolving on a daily basis. 

In addition, multiple developers have revealed that the general availability version of DeepSeek V4 has entered small-scale gray-scale testing and could make an official debut soon. The model currently comes in two variants: V4 Flash and V4 Pro, according to nbd.com.cn. Feedback from testers indicates significant improvements in programming, design, and agent generation capabilities. Some developers believe its overall performance is even on a par with Anthropic's Claude 5, the report noted.

The vigorous development of Chinese LLMs reflects that China is placing greater emphasis on an open and inclusive development path. "Only through openness and inclusiveness can we build a larger and more robust industrial ecosystem, enhance adaptability, and accelerate technological iteration," Hu said.

"If the cost matters a lot, a lot of this may move to China because of the open-source models," Kevin Kelly, the founding executive editor of the US magazine Wired, told the Global Times when asked whether China's open-source AI models could make them more competitive. "Open-source is a fantastic experiment and I'm glad China is going in that direction."

Wang Weiming, an official from the Ministry of Industry and Information Technology (MIIT) said on Monday that China's innovation capability continues to strengthen since the beginning of 2026, with the penetration rate of AI technology across the country's major large and medium-sized enterprises exceeding 30 percent, and the cumulative global downloads of China's open-source AI models surpassing 10 billion.

Hu noted that China is actively promoting open source in order to break the West's closed monopoly in the field of AI. "In the process of developing AI, China places special emphasis on preventing the formation of new digital divides. It advocates for global co-governance and is committed to making AI universally beneficial infrastructure, rather than a tool for profit in the hands of a few developed countries."