Moonshot’s Kimi K3 adds price pressure to top AI models

FORMAT B - Trend

In short: New AI models from China, including Moonshot’s Kimi K3, are getting closer to US leaders and could make advanced AI cheaper.

What's going on

Chinese start-up Moonshot recently launched an AI model called Kimi K3. The Financial Times says it comes close to matching the best models from leading US AI labs.

This follows earlier concern about another Chinese model, DeepSeek R1, which drew attention partly because it seemed cheaper to build. Now, the FT says models such as Kimi K3, GLM 5.2, and Alibaba’s Qwen 3.8 have narrowed the time lead that US companies were seen to have.

Some US officials have accused Moonshot of copying US rivals using a method called distillation (like training a new student by having them closely imitate a top student’s answers). The FT reports that even an OpenAI executive suggested Kimi K3’s progress does not look like simple copying, and that some of its design changes have impressed US observers.

The article also points out a practical detail. Even if a model is cheaper to access, very large models can still be expensive to run. Prices are often based on “tokens” (small chunks of text), but what users really care about is the total cost to finish a job, including time spent checking mistakes.

What to watch

The next big question is how US companies and governments respond. The FT notes talk in Washington about possible crackdowns related to distillation and chip restrictions, and also signals from China about promoting open sharing while considering new limits on exporting key parts of its models. For everyday users, the likely near-term effect is more price competition, plus more uncertainty about which AI tools will be available in which countries.

Source: Financial Times