The World Artificial Intelligence Conference has not even started, and domestic large models have already released a major breakthrough. As the 2026 World Artificial Intelligence Conference and the High-Level Meeting on Global Governance of Artificial Intelligence are about to take place, Moonshot AI officially launched its new model, Kimi K3, on the 16th. With a parameter scale of 2.8 trillion, it has taken the top spot as the largest open-source model in the world. This is not just a numerical achievement, but also marks the first time that the open-source community has surpassed closed-source giants in terms of scale.

Kimi K3 is not just about stacking parameters for show. According to Beijing Moonshot Technology Co., Ltd., this model is natively supported for visual understanding and features an ultra-long context window of 1 million tokens, equivalent to swallowing an entire book and still remembering what was written at the beginning. It has been specifically optimized for complex tasks such as software engineering, knowledge work, deep research, and multimodal understanding, thereby significantly enhancing the ability to handle complex tasks—ranging from image recognition to long-term code engineering. K3 aims to tackle hard problems that previously required multiple steps by integrating them all into a single model.

Capital markets have already reacted. China Securities Investment Research issued a report with confident judgments: global large model usage is expected to rise for 11 consecutive weeks, and domestic models are steadily leading in traffic due to their high cost-effectiveness. Meanwhile, revenue expectations for overseas top manufacturers have also been raised. More intriguing is the shifting commercialization rhythm—domestic companies have clearly accelerated their pace. DeepSeek is preparing to introduce time-based pricing, while Doubao has already launched multiple membership tiers. On the other hand, Anthropic has tightened access restrictions on Claude. Under the combination of compliance and geopolitical factors, the logic of domestic substitution has been further illuminated.

Following this main line, Cailian Press' theme library has identified several listed companies that are resonating with this trend. 360 has focused on AI with its 360 ZhiNao and 360 Security Large Model. Its self-developed 360 ZhiNao, a general-purpose large model with hundreds of billions of parameters, has already achieved capabilities in multimodal understanding, logical reasoning, and code generation. Kunlun Wanwei announced on July 2nd that its TianGong AI achieved a historic breakthrough in the second quarter of 2026, with its AI Native model and product business achieving an annual recurring revenue (ARR) of over $800 million. As the 2.8 trillion parameter open-source giant takes shape, the momentum of domestic large models is being pushed to a visible new peak.