Chinese startup Moonshot on Thursday published its most recent generative artificial intelligence system that purports to outperform OpenAI ChatGPT in “agentic” tasks -or what a user desires but no longer requires step-by-step instructions to do it.
The model is known as the “Kimi K2 Thinking,” and it is an expansion of the K2 model that was launched by the Alibaba-backed Beijing-based Moonshot in July.
The update comes amid this week, Nvidia CEO Jensen Huang once again encouraged the U.S. to continue the race against AI developed by China.
Even some U.S. giants like Airbnb have started publicly boasting of how certain Chinese AI models are as viable, and in many cases, lower-priced substitutes to those of OpenAI.
Although the U.S. has limited access to Chinese businesses to high-end chips, AI models like DeepSeek have been published at a user fee, a fraction of that of ChatGPT.
Meanwhile, the DeepSeek further argued that it incurred an expense of $5.6 million on its V3 model, which is compared to the billions used by OpenAI.
The source familiar with the matter reported that the Kimi K2 Thinking model cost $4.6 million to train.
According to Moonshot, it is capable of automatically picking its 200 to 300 tools to perform tasks independently without human intervention. However, CNBC could not confirm independently the DeepSeek or Kimi figures.
Last month, DeepSeek published a new AI model that says it can enhance its performance by incorporating visual clues to expand the context of the information it is currently processing.
