Chinese Company Moonshot Unveils New Open Source AI Model
Moonshot, a Chinese company, has introduced an open source AI model that resembles those developed by Anthropic and OpenAI. This move highlights China’s advancements in the competitive landscape of AI technology.
The new language model, named Kimi K3, has been trained on an impressive 2.8 trillion parameters, which are the data points influencing how it responds to user queries. If confirmed, Kimi K3 could become one of the largest AI models available.
Founded by Yang Zhilin, Moonshot claims that Kimi K3 has surpassed the performance of Anthropic’s Opus 4.8 and OpenAI’s ChatGPT 5.5 in many coding tasks. However, it still falls short of Anthropic’s advanced Fable model, which was temporarily taken offline in June due to potential misuse concerns.
As an open-source project, Kimi K3 will likely be accessible for download and offer a more affordable alternative compared to the proprietary models from CEOs Dario Amodei (Anthropic) and Sam Altman (OpenAI). For context, Moonshot’s previous model, K2.6, was about one-third the cost of Opus 4.8.
Kimi K3’s announcement follows closely on the heels of Z.AI’s release of GLM-5.2, another open-source model that matches or nearly matches the capabilities of major U.S. counterparts.
Experts are increasingly worried that inexpensive Chinese AI models might challenge the business models of U.S. labs, which often charge hefty fees for the tokens necessary to operate their chatbots.
If Kimi K3 proves effective in real-world applications and not just in testing scenarios, it could disrupt the assumption that the U.S. maintains a 6-12 month advantage over China in developing advanced AI models.
Additionally, Moonshot may face intensified examination from U.S. authorities as concerns mount over China’s alleged use of “distillation,” a method that allows less capable models to learn from more powerful ones without authorization.
Earlier this year, Anthropic accused Moonshot, along with two other Chinese organizations, DeepSeek and Minimax, of using this technique to obtain its AI technology. Furthermore, in June, Anthropic informed Congress about claims that Alibaba had enhanced its models through similar actions.
