Even in Silicon Valley, many of the highest-valued AI startups have built their applications at least in part using Chinese open-source models from providers like DeepSeek, Moonshot AI and Z.ai.
“I don’t think they prefer Chinese models, it’s just the best open-source models are Chinese,” said Lan Xuezhao, a venture capitalist in San Francisco who runs an AI-focused fund called Basis Set. “You go with the best and most efficient and low-cost models.”
On Arena, a popular AI benchmarking platform, the top eight open-source AI models for agent-based tasks are made by Chinese developers, as are the top 17 open-source models for coding tasks.
Thousands of U.S.-based startups, companies and academic researchers are relying on Chinese open-source models, according to Clement Delangue, CEO of the startup Hugging Face, a popular developer platform that hosts open-source models and datasets.
“Any sort of restrictions would be a terrible blow to them and concentrate AI power even more in the hands of a few mega-corporations against whom it would be virtually impossible for anyone to compete,” Delangue told Reuters.
In some use cases, Chinese open-source models are both cheaper than and technically competitive with leading models from the likes of Anthropic, OpenAI and Google. The same cannot be said for the current slate of open-source models coming out of the West, due to the slow release cadence and lagging performance.
OpenAI’s GPT-OSS is nearly a year old — “an eternity in AI timelines,” as Delangue put it — while open-source offerings from both big firms like Nvidia and Google and upstarts like Reflection AI have not kept pace with the advanced capabilities in a model like Z.ai’s recently released GLM-5.2, which had Silicon Valley buzzing with its coding and agent capabilities.
