Developed by the Chinese company Moonshot AI, the new large language model (LLM) Kimi K3 topped one of the key ratings of the Arena portal even before its full release, displacing its American competitors. As the compilers of the top reported, the new artificial intelligence (AI) turned out to be much more effective than Claude Fable 5 and GPT-5.6, at least in the development of program code for user interfaces.
The previous version of the same neural network, Kimi K2.6, was only in 18th position on the chart. The new version contains almost three times more parameters – 2.8 trillion, for which the so-called model weights will be released on July 27. This will make the Kimi K3 the most powerful open source model, meaning other developers will be able to use its architecture for their own purposes.
The updated model also works significantly better than the previous version with text (9th place compared to 38th), but is still inferior to American developments in these tasks. The five first places in this Arena top are held by various versions of LLM Claude from the Californian company Anthropic.
Former senior product manager at Meta and now AI entrepreneur Xiaoyin Qiu believes open weights will give Kimi K3 a major advantage in the AI market. As Futurism notes, even before the release of the Kimi K3, the American market began to take a closer look at Chinese developments, not wanting to “pay six times more” for American ones.
“When the best open weight model beats the best closed source model, how can Anthropic justify its prices on [Claude] Fable? Why would anyone pay for this?” Qiu noted on X (formerly Twitter).
She also recalled the recent scandal with the US government's demand to limit access to advanced versions of Claude due to security concerns. According to Qiu, Chinese developers are not worried about this problem, and they have also demonstrated the ability to achieve results in creating AI in the face of restrictions on the supply of American Nvidia chips.
The publication of the full version of Kimi K3 on July 27 can now become an important milestone for investors, writes the media platform Crypto Briefing. Once developers are able to run and test the Kimi K3 themselves, it will become clear whether the stated characteristics correspond to reality.
In 2025, the release of the Chinese AI DeepSeek led to the loss of seven US technology giants $1 trillion on the stock exchange, as RTVI wrote. Subsequently, American leaders in AI, including Google and OpenAI, managed to reassure investors with the release of more powerful, cutting-edge models.
