China is rapidly emerging as a major power in the AI world. Following DeepSeek’s success, Moonshot AI, a Chinese AI startup, has now introduced a new model called Kimi K2, which is creating buzz. As per the reports, this model is huge, spanning 1 trillion parameters and even surpasses OpenAI’s GPT-4 in some focal points, reported Money Control.
Kimi K2 is not just tasked to chat with users. As per Moonshot AI, the model is capable of ‘acting’ – meaning, it is equipped to use tools, write, program, run code, and accomplish complex tasks on its own without any step-by-step guidance.
Kimi K2 comes with two versions – while one is a foundation model built primarily for researchers and developers, the latter is designed for casual uses like digital assistants and chatbots.
In benchmark tests, Kimi K2 is knocking out!
For instance, it scored a 65.8% in a difficult software engineering test, beating other open-source models out there.
Surprisingly, it outperformed even DeepSeek and GPT-4.1 in coding tests, scoring 53.7% on LiveCodeBench, while GPT-4.1 could score only 44.7%. In Math, Kimi K2 shocked everyone with a record of 97.4% accuracy – again beating GPT-4.1, which scored a 92.4%.
The expenses involved in training the model have also left everyone staggered after Moonshot AI claimed that they have trained Kimi K2 using less money and computing power compared to other AI giants like OpenAI.
While major tech giants pour hundreds of millions into training, Moonshot appears to have discovered a more cost-effective, faster, and smarter approach to developing cutting-edge AI.
