Open models are overtaking AI’s front line

While attention this summer was fixed on Anthropic’s newest frontier models and Washington’s fight over access, developers kept moving in another direction. On Hugging Face, Chinese open-weight models accounted for 41% of downloads this spring, surpassing U.S. models. On OpenRouter, the top six most popular models are all open models from Chinese firms including Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. Anthropic’s Claude Opus 4.7 sat in seventh place at the time of writing.

Other platform data points the same way. Vercel says open-weight models are taking on much of the high-volume infrastructure behind AI apps, while closed models remain the pricier premium tier. In June, open models handled nearly a third of AI requests on the platform. Those figures cover only part of the market and exclude sessions hosted directly by major labs, where OpenAI and Anthropic likely still see the bulk of their usage.

For Hugging Face CEO Clem Delangue, the trend suggests frontier systems may end up reserved for a narrower set of jobs.

“Maybe in a few years, the frontier models will be for experimenting and [for] some really high-value tasks, and most of the production workloads will actually be powered either by private models within companies or by open source models.”

Delangue said customers increasingly want to own models rather than rent access through black-box APIs, especially after seeing what it costs to scale closed frontier systems. He argues companies do not want to outsource core capabilities to providers they cannot inspect or control.

Recommended reading

1010Benja made a Suno song that actually works

That shift is visible on Hugging Face itself. According to Delangue, the platform sees a new repository created every seven seconds and hosts almost three million public models and one million public datasets. Half of all Fortune 500 firms are using Hugging Face to deploy private and open source models, he said. The result, in his view, is not “one model to rule them all” but a market made up of many specialized models tuned for specific use cases.

Chinese open-weight releases keep pressuring U.S. rivals

The rise of open models has coincided with a steady stream of stronger releases from Chinese AI labs. Every few months, another company ships an open-weight model that is cheaper to deploy and easier to customize than closed alternatives, putting pressure on the economics behind proprietary AI systems.

Most recently, Beijing-based Z.ai released GLM-5.2, an open-weight model that, according to the source, performs strongly at agentic coding and competes with Anthropic’s latest models on identifying security vulnerabilities.

Microsoft CEO Satya Nadella has made a similar case against relying on a single provider, framing control over data as a central concern for enterprises.

“While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation, and to reserve the right to learn from customer usage and interaction data.”

“If learning flows in only one direction, economic value converges toward the owners of the learning infrastructure rather than the creators of the knowledge itself. Therefore, it’s imperative that we distribute the learning infrastructure to every firm so that they can control their own learning loop.”

Open-model safety debate is getting sharper

The momentum behind open models is also deepening the fight over whether highly capable systems should be widely released at all. Anthropic CEO Dario Amodei has argued that powerful open model weights could become dangerous because once released, they are hard to control. Critics of open models have also warned that bad actors could use them for disinformation, cyberattacks, or biological warfare.

Delangue argues the opposite risk matters more: concentration of power.

“The biggest risk in AI is concentration of power.”

He says transparency helps defenders understand and patch known cybersecurity risks, while keeping advanced models closed does not remove the underlying dangers. In his view, API guardrails can be bypassed and model weights can still be stolen and distributed. Restricting access, he argues, mainly leaves powerful systems in the hands of a small number of companies.

“You don’t really make it safe by keeping it behind closed doors for just a few players. You make it more dangerous because you create asymmetry of power and asymmetry of capabilities.”