The Chinese startup Moonshot has introduced the artificial intelligence model Kimi K3. According to the company, this system is capable of processing a significantly larger volume of data per query.
Kimi K3 was developed to perform tasks requiring advanced thinking, complex programming, and intensive operations.
Arthur Igreya explained how Moonshot's new product, Kimi K3, promises to raise the level of artificial intelligence development.
The Chinese startup Moonshot has actively entered the competition for leadership in artificial intelligence with the launch of Kimi K3. The company claims that this open-weights model has 2.8 trillion parameters, positioning it as the largest of its kind in the world.
Kimi K3 was developed with a focus on performing tasks that require sophisticated reasoning, long-term programming, and knowledge-intensive activities. According to Moonshot, it represents the first open-weights model to approach the three-trillion parameter mark. In addition to its size, the model features a context window of 1 million tokens, allowing it to process a significantly larger amount of data in a single request compared to previous versions.
The main attributes highlighted by the company include: 2.8 trillion parameters; a 1 million token context window; focus on complex programming and advanced reasoning; and being an open-weights model, which allows for customization by users. This feature enables developers to download, run, and fine-tune the system for various applications, unlike closed models.
Moonshot reported that Kimi K3 demonstrated competitive performance when compared to Fable 5 and surpassed Opus 4.8, GPT 5.6 Sol, and GPT 5.5 in GPU kernel optimization. This optimization is a technique used to improve hardware utilization and reduce latency.
The results were corroborated by external evaluations. The Arena.ai platform ranked the model in the first position in a specific test on web interface creation. Vals AI ranked it second, only behind Fable 5 but ahead of GPT-5.6 Sol. In sequential complex task tests, Artificial Analysis found performance similar to GPT-5.5 and Claude Opus 4.8.
Despite the progress, there are limitations to consider. Lian Jye Su, chief analyst at Omdia, mentioned that the execution cost could be a fraction of the price charged by OpenAI to its clients. However, he stressed that this does not automatically guarantee the best performance. On the other hand, Ryan Fedasiuk, a researcher at the American Enterprise Institute, warned that running a model of this magnitude locally would require equipment estimated to cost hundreds of thousands of dollars.
Reuters reported that Moonshot, supported by Alibaba and Tencent, continues to expand its financial and technological capacity. As reported by Bloomberg, the startup is seeking to raise $2 billion (equivalent to about R$ 11.1 billion), with a valuation close to $30 billion (approximately R$ 166.5 billion) before a potential public offering in Hong Kong. With the launch of Kimi K3, the company strengthens its strategy to compete in the global artificial intelligence landscape during this period of rapid industry evolution.
Last Friday, Xi Jinping positioned Beijing as a defender of a new world order in the field of artificial intelligence. During a prestigious Chinese technology conference, he actively promoted open-source technologies and challenged US influence regarding the regulation of this rapidly evolving industry.
Speaking at the opening ceremony of the World Artificial Intelligence Conference (WAIC) in Shanghai, Xi called on countries to take advantage of the 'rare and historic opportunity' of open AI. He also pledged support to developing nations in advancing AI capabilities, warning against the threat of 'new historical injustices' due to unequal access to this technology.
These statements represented the clearest articulation of China's ambitions to shape global AI governance. Xi presented his open-source models as a public good for the entire world, positioning Beijing as an alternative to Washington at a critical juncture in the race for technological leadership.
Comparing the significance of AI to the invention of the steam engine and electricity, Xi outlined his vision wherein China would share AI technologies and experience with Global South nations while leading global efforts to establish standards for this new technology. His initiative with the AI coalition was presented as a competitor to the international initiative led by the US, known as 'Pax Silica,' aimed at securing global AI supply chains and critical minerals, although Xi avoided direct mention of Washington.
Xi's comments came amid the rapid progress of Chinese open-weight AI models against the proprietary systems of American companies such as OpenAI and Anthropic. On Friday, the Beijing-based startup Moonshot AI unveiled Kimi K3, which was named the largest open-source AI model by parameter count. This occurred a month after the US government abruptly withdrew advanced Anthropic AI models due to safety concerns.
Furthermore, Xi insisted that AI systems must remain under human control, urging countries to create early warning and emergency response mechanisms to manage AI risks. He also emphasized the need for measures to prevent loss-of-control scenarios, warning about the dangers of autonomous AI systems that might evade human oversight and control.
Xi stated that the China AI Cooperation Organization, established in China and adopted by 29 member states on Thursday, marks a 'milestone in the development of global AI' and meets the demands of Global South countries for more active participation in AI governance. Moreover, Xi announced that China would provide AI training and develop AI cooperation centers with BRICS, ASEAN, Latin America, and the African Union countries, thereby aligning China's diplomatic initiatives in AI with major Global South blocs where China already holds significant influence.
George Chen, Head of Digital Practice at the consulting firm The Asia Group, noted: 'Xi's message is clear: China does not intend to follow anyone, neither in AI technologies nor in standards. Instead, China intends to lead the world in both aspects.' He added that Xi's message is a statement that can also be viewed as a warning—China will not allow anyone to dictate how it uses AI.
The event, running from July 17 to 20, takes place against the backdrop of Washington and Beijing preparing for the first bilateral government-level negotiations on AI under the administration of US President Donald Trump. This transforms WAIC from an industry exhibition into an early test of how China plans to compete for influence in global AI regulatory rules. Last week, the two powers presented competing concepts during the UN AI dialogue, where US representatives argued that excessive regulation could slow innovation, while China promoted its low-cost open-source models as a tool to reduce global inequality in AI access.
