The new artificial intelligence model "Kimi Kai 3" launched by Chinese startup "Moonshot AI" on Friday has garnered widespread admiration due to its capabilities approaching the performance of the most advanced American AI systems, prompting calls in the United States to move quickly to keep pace with this development.

Since the release of DeepSeek's "V1" tool in December 2024, Chinese AI companies have been continuously reshaping the sector with open-source models that can be downloaded and modified, as well as being available for free.

Related news

These models partially reconsider the economic foundations of AI established by major Western companies, which are based on paid and closed models.

Adjustable parameters

With the "Kimi Kai 3" model, a new level has been reached, especially in terms of size, as it was created using 2.8 trillion parameters (adjustable parameters), about twice the size of "DeepSeek V4 Pro" launched in April with 1.6 trillion parameters. Moonshot AI boasts that its new model represents "the upper limit in terms of size of open models" in the field of AI. But what caused surprise above all is the performance of "Kimi Kai 3", as AI models are now compared according to a wide range of specific benchmarks and tasks.

AI models

In this race, "Kimi Kai 3" approaches advanced AI models like Anthropic's "Fable 5" and OpenAI's "GPT 5.6 Sol" in many aspects. In some rankings, Moonshot AI's tool topped the list, especially in the field of application or website programming, according to the ranking by the reference platform "Arena AI". At the current stage, code generation is the most prominent use of generative AI, a market currently dominated by Anthropic, while OpenAI continues to strengthen its presence at an accelerating pace. Alex Finn, head of the "Henry Intelligent Machines BBC" platform, said in a post on X platform that this "will bring about a radical and permanent transformation in the AI race."