- Chinese AI startup Moonshot has launched Kimi K3, a 2.8 trillion-parameter open-weight model.
- The release triggered sharp falls in the shares of rival Chinese AI companies as investors reacted.
- Analysts say the launch highlights China's rapid progress in AI, though questions remain over real-world performance.
China's artificial intelligence race has taken another step forward with the launch of Kimi K3, a new large language model that its developer claims can compete with some of the most advanced AI systems from the US.
Beijing-based startup Moonshot AI unveiled the 2.8 trillion-parameter model on July 18, describing it as the world's largest open-weight AI model. Unlike closed-source systems, open-weight models allow developers to download, customise and run the software on their own infrastructure.
The launch immediately caught investors' attention. Shares of Chinese AI firms Zhipu and MiniMax fell 27.7 per cent and 16.5 per cent respectively in Hong Kong, reflecting expectations that competition in China's AI market is becoming even more intense.
A bigger model, but does bigger mean better?
Moonshot says Kimi K3 is the first open-weight model to approach the 3 trillion-parameter mark and has been built for demanding tasks such as software development, complex reasoning and processing lengthy documents through its one million-token context window.
The company also claims the model performs at a level close to Anthropic's latest Fable system while outperforming several leading models in areas such as GPU optimisation and long-horizon coding. Those claims have yet to be independently verified across every benchmark.
External evaluations, however, suggest Kimi K3 is already performing strongly. Arena ranked it first for web interface building, while Vals AI placed it second overall behind Anthropic's Fable 5 and ahead of several OpenAI models. Artificial Analysis also found its performance broadly comparable with leading commercial AI systems on complex reasoning tasks, according to Reuters.
Industry experts say the launch reflects how quickly Chinese developers are narrowing the gap with their US rivals.
Omdia chief analyst Lian Jye Su reportedly said Chinese AI models can often be deployed at a fraction of the cost charged by US companies. However, he added that having a larger model does not automatically guarantee better performance.
Hussein Abbass, a computing professor at UNSW Canberra, reportedly said Kimi K3 appears particularly strong at coding but that its performance across the full range of foundation model tasks still needs further evaluation.
China's AI challenge grows louder
Kimi K3 arrives just weeks after the US government halted the public release of Anthropic's Fable and Mythos models over security concerns, adding another twist to an increasingly competitive AI landscape.
The launch also follows a string of high-profile Chinese releases, including DeepSeek and Z.ai's GLM-5.2, both of which challenged the long-held view that Chinese AI developers were significantly behind their US counterparts.
AI researcher Ethan Mollick reportedly described Kimi K3 as "closest to the frontier yet", while noting that even the latest models continue to struggle with certain creative tasks.
Despite the excitement surrounding the release, experts caution that model size is only one part of the picture. Training data, computing infrastructure, specialised chips and software optimisation all play an important role in determining how capable an AI system ultimately becomes.
Another practical challenge is cost. Ryan Fedasiuk of the American Enterprise Institute reportedly said running a model of Kimi K3's size locally would require computing equipment worth hundreds of thousands of pounds, limiting its accessibility despite its open-weight design.
Even so, the launch adds to growing evidence that China's AI companies are moving quickly, not only improving performance but also offering powerful models at lower prices. That combination could increase pressure on leading US AI firms as competition in the global market intensifies.
