New Chinese AI model Kimi K3 challenges OpenAI and Anthropic
The US technology sector was caught off guard last Friday by the launch of yet another advanced AI system from China. The Kimi K3 model, developed by Chinese start-up Moonshot, showcases capabilities that can rival American frontrunners such as OpenAI’s ChatGPT and Claude by Anthropic.
According to Anastasios Angelopoulos, CEO of AI evaluation site Arena, this launch could well be the most important of the year. It points to a shift in which Chinese open source technology is increasingly competing with the closed AI systems of American companies. According to the Associated Press, K3 now tops the Arena rankings for front-end coding.
The timing of the launch coincided with President Xi Jinping’s opening speech at the annual World Artificial Intelligence Conference in Shanghai. Although US-imposed trade barriers have limited China’s access to high-end technology, these restrictions have in fact pushed the country to accelerate its own domestic innovations.
During the summit, President Xi said that progress in the field of AI should be a joint global effort rather than a competition dominated by a single country.
The development fits into a broader trend of rapid advances in the Chinese AI sector. Only last month, start-up Zhipu released GLM-5.2, which is gaining popularity among developers worldwide because it delivers performance almost equal to that of American models at lower cost.
Although some analysts, such as Patrick Moorhead, suggest that the current excitement around K3 is an overreaction, similar to the panic following the release of DeepSeek, he notes that such advances represent a significant threat to the revenues of companies like OpenAI and Anthropic.
Hardware independence is also becoming a priority for China. Huawei recently presented the Atlas 950 SuperPoD, with which the company demonstrated its own AI computing capabilities despite US restrictions on Nvidia chips.
Although Moonshot has not disclosed which specific hardware was used to train K3, the company maintains a partnership with Huawei. Analysts at Bank of America further report that K3’s costs are significantly lower, roughly half those of OpenAI’s advanced GPT-5.6 Sol model.
The rise of Chinese AI models has also led to accusations of so-called ‘distillation’ where companies train smaller models based on the outputs of more powerful systems in order to avoid high research and development costs.
Anthropic has accused Moonshot, DeepSeek and MiniMax, among others, of using this method to appropriate technology. China denies those accusations. At the same time, technological exchange runs in both directions; San Francisco start-up Anysphere admitted that its programming tool, Cursor, made use of Moonshot’s K2.5 model.
Moonshot was founded by Yang Zhilin, an alumnus of Carnegie Mellon University and former student of Russ Salakhutdinov. Zhilin’s academic contributions to machine learning have earned him the respect of former colleagues and mentors, who see his success as a victory for the open source movement.
Although open source development is praised for speeding up innovation by making technology transparent and adaptable, critics continue to warn that it can bring new security and safety risks.
Business AM
Subscribers 0
Fans 0
Followers 0
Followers