After DeepSeek, China launches Kimi K2 AI model that outperforms GPT-4 in benchmarking tests

ChatGPT

China is quickly becoming a major force in the AI world. Just after DeepSeek’s success, a Chinese AI startup called Moonshot AI has launched a new model named Kimi K2, and it’s already making waves. According to reports, this model is not just big, it’s huge, packing 1 trillion parameters and it even beats OpenAI’s GPT-4 in some key areas.

What’s so special about Kimi K2? Well, it’s not just designed to chat with users. Moonshot AI says this model can actually “act”, meaning it can use tools, write and run code, and carry out complicated tasks on its own without anyone having to guide it step by step.

There are two versions of Kimi K2: one is a foundation model mainly for researchers and developers, and the other is tuned for more casual uses like chatbots and digital assistants.

In benchmark tests, Kimi K2 is scoring big. For example, it got 65.8% on a difficult software engineering test, ahead of most open-source models. It also did better than DeepSeek and GPT-4.1 in coding tests, scoring 53.7% on LiveCodeBench, while GPT-4.1 only managed 44.7%. When it comes to math, Kimi K2 blew everyone away with 97.4% accuracy, higher than GPT-4.1’s 92.4%.

What’s more interesting is the cost. Moonshot AI claims they’ve trained Kimi K2 using much less money and computing power compared to companies like OpenAI. While big tech companies spend hundreds of millions of dollars on training, Moonshot seems to have found a cheaper, faster, and smarter way to build top-notch AI.

With Kimi K2, China is showing it’s not just catching up in AI, it’s actually pulling ahead in some areas. The global AI race just got a whole lot more interesting.