Chinese AI Startup Releases Massive Open Weight Model

Allexxandar via Getty ImagesChinese AI vendor Moonshot AI on Friday released the latest version of its flagship open weight model, further provoking discussion about the availability of open models from Chinese vendors despite the geopolitical tension between the U.S. and China.Kimi K3 is a massive multimodal AI model that rivals Anthropic's Claude Fable 5 and even OpenAI's GPT Sol. The model seems to be the largest open weight model to date and more cost-effective than proprietary models from U.S. vendors.The model's introduction highlights the conundrum faced by enterprises interested in open models. Despite the strength of proprietary models from OpenAI, Anthropic and Google, the U.S. open market is not as strong as China's, and enterprises want options."For enterprises serious about training their own models, optimizing their own models, it does feel like if you look beyond Nemotron, Nvidia's offering, it's all Chinese now," said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget. "The question is, how do they navigate that situation? There is an argument to be made that it is fine for companies to have a diverse collection of models and use them according to their specific use case."Related:Thinking Machines Rolls Out Broad but Efficient ModelOpen Models in the U.S.For some enterprises, the tension between the U.S. and China might be a hindering factor to using a model like Kimi K3. While not new, the push-pull between the two countries continues to evolve, from the U.S. applying export controls to keep China to more targeted policy controls and retaliations from China, which placed its own restrictions on U.S. chips.Without geopolitical tension between the two countries, open models such as DeepSeek and Alibaba Qwen would likely be broadly embraced by U.S. businesses, Su said."From an enterprise standpoint, it's more about the cost of getting work done, and then if a model can deliver the right value, there is no reason why an enterprise should say no or should always choose or go with a proprietary model," Su said.For U.S. model makers, the tension and the growing open source market in China could signal an opportunity to create strong open models for enterprises. Some vendors are taking advantage of the opportunity.Earlier this week, AI startup Thinking Machines introduced its new open weight model Inkling, aimed at competing with its proprietary counterparts. IBM and research vendor AI2 are also focused on open weight models. Even so, open models from the U.S. trail behind those offered by Chinese vendors.Kimi K3 proves scale mattersWhile the U.S. market plays catch-up, Chinese vendors such as Moonshot are revising their techniques to make their models even stronger. Kimi K3 has almost 3 trillion parameters, a move that Su said proves that scaling laws still matter in the AI industry.Related:Anthropic Teams With Wall Street Firms on AI Venture"We are still seeing models getting larger, and these models are the smartest in the industry, the most powerful in the market," he said.He added that, beyond parameter size, techniques matter too. Kimi K3 utilizes a mixture-of-experts architecture, which means not all parameters are activated at the same time. The model is also optimized for agentic coding rather than for general-purpose use."We are in an era where there are different ways to train an AI model, and the question is about your use case and your purpose, optimally," Su said.About the AuthorNews Writer, AI BusinessEsther Shittu has covered AI technologies and industry trends since 2021. As co-host of the Targeting AI podcast, she talks with experts, thought leaders and practitioners exploring critical AI developments. Before AI Business, she wrote for SearchEnterpriseAI, the New York Daily News, Bklyner and the Brooklyn Daily Eagle. When she's not diving deep into the world of AI, she spends her time on passion projects and raising her three daughters.