The launch of the Kimi K3 AI model by Moonshot, an early-stage Chinese AI company, is yet another reminder, as if one were needed, that the AI race between the U.S. and China is no foregone conclusion.
Kimi K3 has 2.8 trillion parameters.
Well, that sounds like a lot of something.
Parameters?
Parameters in AI refer to the variables that are used to define and manipulate the input and output of machine learning models. These variables can include weights, biases, and other factors that are used to adjust the behavior of the model. Parameters are an important aspect of machine learning models, as they help to determine the accuracy and reliability of the model.
To oversimplify, they are (in a way) a model’s “settings.”
According to Zijing Wu and Jamie John, writing in the Financial Times, the higher the count, the greater the model’s capability. K3 has a higher count than Anthropic’s flagship Claude Opus 4.8. It’s also thought that it will be open-weight, which is to say, downloadable and able to be modified by those using it, unlike the most advanced U.S. models.
Wu and John’s description of the Chinese approach to the market sounds . . . familiar:
Companies from Silicon Valley to Europe are also switching to cheaper Chinese models to reduce their rising bills for technology from US labs.
The US’s top model makers have poured hundreds of billions of dollars into building out infrastructure and developing the most advanced AI tools. They have begun to charge steeper fees for companies to access them this year. Anthropic will increase the price of Opus 4.8 by 50 per cent to $3 per million input tokens and $15 per million output tokens in September, according to its website.
Chinese labs such as Moonshot and DeepSeek, meanwhile, have released open-weight models that are cheaper to run and can be downloaded and modified by users. Moonshot’s K2.6 model, for instance, costs about a third of Anthropic’s Opus 4.8.
The two also note the perception that, at the highest end, the gap between Chinese and U.S. companies is closing.
Bloomberg’s Sunny Bangia notes one consolation for chipmakers. When DeepSeek’s R1 was released last year, chip stocks went into a spin, hit by concerns that DeepSeek needed more computing power. K3 had a similar effect on Friday, but:
DeepSeek’s breakthrough centered on making AI models cheaper to train and run. Kimi K3 also improves computing efficiency, but does so with a far larger model that places heavier demands on memory infrastructure. That could continue to support the need for high-bandwidth memory from SK Hynix Inc., Nvidia Corp.’s latest AI systems and advanced chipmaking from Taiwan Semiconductor Manufacturing Co.
Let’s see.
Meanwhile, this news only reinforces the importance of not slowing the development of American AI either by zealous regulation or by subjecting it to international “governance,” two mechanisms that Beijing will exploit with zeal.