Artificial Intelligence helps police recover vehicles and arrest suspects Chinese artificial intelligence startup Moonshot launched this Friday (17) Kimi K3, a new AI model that, according to the company, is the largest open source system in the world. The technology has 2.8 trillion parameters - an indicator of the size and capacity of the model - and, according to the company, delivers performance close to that of Fable, one of the most advanced systems from the North American company Anthropic. The launch comes about a month after the Fable and Mythos models, also from Anthropic, were suspended by the United States government due to security concerns. The episode reinforces the speed with which Chinese companies have been closing the gap with American leaders in the development of artificial intelligence. In addition to Moonshot, companies like Z.ai and MiniMax have also presented increasingly powerful and lower-cost models. This advance challenges the idea, common until recently, that China was many months behind the United States in the AI ??race. According to Moonshot, the Kimi K3 is the first open source model to approach the 3 trillion parameter mark. It is designed to handle complex tasks such as programming, problem solving and analyzing large volumes of information. The system can also process a much larger amount of text at once than previous generations. Unlike closed models, which can only be used through the platforms of the companies that create them, open source models can be downloaded, executed and adapted by users themselves. Read more: AI can transform the economy faster than the Industrial Revolution, experts warn Owner of ChatGPT launches R$1,000 keyboard to control AI agents; see how it works The AI ??race raises expectations around the tools and the millions of dollars they can generate. Dado Ruvic/Reuters/Illustration The company claims that the Kimi K3 achieved performance similar to that of Fable 5 and surpassed models such as the Opus 4.8, from Anthropic, in addition to the GPT-5.6 Sol and the GPT-5.5 in tests that measure the efficiency in the use of the hardware responsible for processing the AI, which helps to make responses faster. The new model also obtained good results in independent evaluations. At Arena.ai, it came first in a test aimed at creating interfaces for websites. Vals AI placed it in second place overall, behind only Fable 5 and ahead of GPT-5.6 Sol. Artificial Analysis stated that Kimi K3's performance is comparable to that of GPT-5.5, from OpenAI, and Claude Opus 4.8, from Anthropic, especially in more complex tasks that require several steps. The announcement also affected the financial market. Shares of Chinese competitors Zhipu and MiniMax fell 27.7% and 16.5%, respectively, on the Hong Kong Stock Exchange. Artificial intelligence gains ground as a competitive differentiator in companies Artificial Intelligence Race for AI accelerates Chinese companies have reduced the time between one launch and another of new AI models. A few days earlier, Z.ai presented the GLM-5.2, which drew attention for achieving results close to those of the main American systems in performance tests. For Lian Jye Su, chief analyst at consultancy Omdia, one of the main differentiators of Chinese companies is the cost. According to him, these models can be used for a fraction of the price charged by OpenAI. Still, the expert emphasizes that a larger model is not always automatically the best. Furthermore, the size of the Kimi K3 makes it difficult for users to run it on their own computers. Ryan Fedasiuk, a researcher at the American Enterprise Institute, said running a system with 2.8 trillion parameters requires equipment costing hundreds of thousands of dollars. What does it mean to have trillions of parameters? Parameters are the values ??that an artificial intelligence model adjusts during training to learn how to perform tasks. In general, the greater the number of parameters, the greater the system capacity tends to be, although this does not guarantee, in itself, superior performance. Before the Kimi K3, the largest Chinese models were Meituan's LongCat-2.0 and DeepSeek's V4-Pro, both with around 1.6 trillion parameters. American companies such as OpenAI and Anthropic do not disclose the number of parameters of their most advanced models, which makes direct comparisons difficult. Moonshot claims that the Kimi K3 brings improvements that increase its efficiency and allow it to perform programming tasks with little human intervention. With support from investors such as Alibaba and Tencent, the company has been expanding its investments to compete for leadership in the global artificial intelligence market.