Moonshot's Kimi K3 pushes boundaries of AI
By Li Jiaying | chinadaily.com.cn | Updated: 2026-07-18 09:10
Chinese artificial intelligence startup Moonshot AI unveiled Kimi K3 on Friday, a 2.8-trillion-parameter model that pairs native vision with a 1-million-token context window, in an indication of the resources being targeted at the area by developers in China.
Moonshot describes K3 as the world's first open model in the 3-trillion-parameter class. The model is already available through Kimi's consumer, workplace, coding and application programming interface services. The company said it will release the full weights by July 27, a step that could give researchers and businesses access to an open-weight system of unprecedented scale.
K3 is aimed at long-horizon work rather than simple question answering. According to developers, it can process large software repositories, operate terminal tools and combine coding with visual feedback. Native image and video understanding also allows it to revise interfaces, games and other digital content after inspecting its own output.
Industry benchmark tables have placed K3 among the world's strongest systems. Independent evaluator Artificial Analysis gave K3 an Intelligence Index score of 57, placing it third globally behind only Claude Fable 5 and GPT-5.6 Sol.
K3's premium pricing is also a notable break with the low-cost strategy associated with many Chinese models. Its API charges 100 yuan ($14.7) per million output tokens, compared with 6 yuan for both DeepSeek V4-Pro and Xiaomi's MiMo-V2.5-Pro, 28 yuan for Z.ai's GLM-5.2 and 36 yuan for Alibaba's Qwen3.7-Max.
According to a Morgan Stanley report, K3's premium pricing strategy could have a far-reaching positive impact on China’s large language model market. China's LLM sector has long been marked by intense price competition, and K3's pricing could help steer the industry toward more sustainable business models while narrowing the pricing gap between Chinese models and those in the United States, it said.
The report also noted that the launch signaled that Chinese LLMs were catching up with US leaders in model scale, performance and pricing, with China expected to produce more models that are larger, higher-priced and more capable and competitive.





















