Chinese AI model Kimi K3 rivals OpenAI, Claude at half the price

On Friday, Moonshot, a Beijing-based AI startup, unveiled Kimi K3, a new large language model that has drawn comparisons to OpenAI's ChatGPT and Anthropic's Claude. The release was announced shortly before Chinese President Xi Jinping's address to the World Artificial Intelligence Conference in Shanghai, where he emphasized that AI development should be a "symphony of global cooperation" rather than a "solo performance by any single country."

According to Arena, a platform for evaluating AI systems, Kimi K3 topped the charts in the ranking of "front-end coding capability," a measure of an AI model's performance on software development tasks. Anastasios Angelopoulos, co-founder and CEO of Arena, called it "the single biggest release of the year" and noted that open-source Chinese models are now surpassing closed U.S. models. He stated on social media that "more results are rolling in that are likely to continue to show it is at the top of the pack."

The pricing advantage is significant: K3 is half as expensive as OpenAI's GPT-5.6 Sol model, according to a Friday report by Bank of America research analysts. Though K3 carries the highest price tag of any Chinese AI model released to date, the cost differential against U.S. offerings remains substantial. This release follows GLM-5.2, a flagship model from Chinese startup Zhipu released last month, which software developers report can perform work almost as well as top U.S. models at a lower price. The market reaction echoes the disruption caused by DeepSeek's model release in early 2025, though some analysts view it differently—tech analyst Patrick Moorhead described the K3 response as an "overreaction shockingly similar" to DeepSeek's release but acknowledged it could pose a revenue challenge to Anthropic and OpenAI.

Moonshot is led by CEO Yang Zhilin, who earned his Ph.D. in 2019 at Carnegie Mellon University. Zhilin was known at Carnegie Mellon for his love of rock bands like Pink Floyd and for making fundamental contributions to machine learning. His former adviser Russ Salakhutdinov, now retired from Apple where he directed AI research, publicly celebrated the achievement, writing on social media: "What a huge win for the open-source community! It feels like just yesterday Zhilin was graduating from my lab at CMU." Moonshot has not disclosed what hardware it used to build K3, but the startup is a partner with Huawei, which is using the Shanghai conference to showcase a new AI computing system called the Atlas 950 SuperPoD—a signal that China is building domestic hardware capabilities despite U.S. restrictions on chipmaker exports.

The release has reignited a dispute over AI model distillation. In February, Anthropic accused DeepSeek, Moonshot, and MiniMax of engaging in campaigns to "illicitly extract Claude's capabilities to improve their own models" using the distillation technique, which involves training a less capable model on the outputs of a stronger one. Anthropic acknowledged that distillation can be a legitimate training method but argued it becomes problematic when competitors "use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently." Beijing has labeled the accusation "groundless." Complicating the narrative, San Francisco startup Anysphere, maker of the popular coding tool Cursor, acknowledged that one of its top products was based on Moonshot's K2.5 model. Elon Musk's SpaceX is planning to close a deal to buy Cursor for $60 billion（約9.6兆円） later this year.

Kimi K3 represents the latest in a series of powerful AI releases from Chinese startups that are narrowing the gap with U.S. tech leaders. The model's unveiling came shortly before Chinese President Xi Jinping's address to the World Artificial Intelligence Conference in Shanghai, where he stated that "the development of artificial intelligence should not be a solo performance by any single country but rather a symphony of global cooperation." This positioning reflects China's broader strategy to build domestic AI capabilities in response to U.S. restrictions blocking access to advanced technologies.

The competitive landscape has shifted noticeably. Last month, Zhipu released GLM-5.2, which developers say performs nearly as well as top U.S. models at lower cost. DeepSeek's model release in early 2025 sparked similar market concerns. While some analysts, including Patrick Moorhead, characterize the K3 response as an "overreaction," the consensus among evaluators like Arena's Angelopoulos is that these releases genuinely represent a meaningful performance leap. The price advantage—K3 costs half as much as OpenAI's GPT-5.6 Sol—creates tangible commercial pressure for Anthropic and OpenAI.

Underlying this competition is the distillation controversy: Anthropic has alleged that Chinese labs, including Moonshot, used the technique to extract Claude's capabilities faster and more cheaply than independent development would permit. The irony is not lost—Anysphere, a San Francisco startup, built one of its own products on Moonshot's K2.5 model, showing that capability transfer flows in multiple directions. Moonshot's partnership with Huawei, which is simultaneously showcasing its Atlas 950 SuperPoD hardware system at the Shanghai conference, suggests China is building both software and hardware stack independence despite U.S. export controls.

No comments yet. Be the first to share your thoughts!