Two weeks after the US government forced Anthropic to pull its two most powerful models off the market worldwide, the early evidence shows export controls are not slowing rivals so much as redirecting the race — and the shift is visible on public charts. Open models are surging, community efforts are openly reconstructing the banned capabilities, and the ban has handed its strongest marketing to exactly the downloadable models the controls were meant to keep ahead of.
Key facts
- What: Two weeks after the US pulled its top models off the market, a Chinese open model sits atop the global download charts and the community is busy rebuilding the banned capability in the open.
- When: 2026-06-24
- Primary source: read the source
The most visible sign is GLM-5.2, an enormous open model from the Chinese lab Z.ai, which now sits at or near the top of the global trending list on Hugging Face, the main public hub where AI models are shared. We covered GLM-5.2 when it launched, in the story of an open model taking on the giants; the new development is not the launch but the momentum. It is released under a permissive license with no regional restrictions — anyone, anywhere, can download it, run it, and build on it, with no government able to switch it off. In a month where the headline lesson was that a hosted American model can vanish on a government memo, a frontier-grade model that physically lives on your own hard drives is a very different value proposition.
That is the heart of the dynamic. When the US pulled its flagship models, it did not just remove two products; it underlined a risk that businesses had mostly ignored — that depending on a single hosted provider is fragile, because the provider, or a regulator standing behind it, can cut you off. The natural hedge is a model you control outright, which is why we have argued that open weights have quietly become a kind of insurance policy. The ban handed the strongest possible marketing to exactly the open, downloadable models the controls were partly meant to keep ahead of. To understand why this category matters so much right now, our primer on open-weight models lays out the trade-offs.
There is a second, stranger signal lower down the same charts. Among the most-downloaded and most-remixed models right now is a cluster of community fine-tunes openly attempting to reconstruct the capabilities of the very models the government just restricted — amateur and semi-professional efforts to distill, approximate, and rebuild the banned models' strengths in the open, where no directive can reach them. The intent is clear and it is a direct, almost gleeful response to the ban: you can pull a product, but you cannot easily pull an idea once thousands of people have decided to chase it.
This is what an export control looks like when it collides with an open ecosystem. The point of restricting a capability is to deny it to rivals. But capabilities are not only embodied in specific products — they are embodied in published research, in open weights, and in a global community of people racing to reproduce whatever is hot. Restrict the product, and you can accelerate the open alternatives and motivate the reconstruction effort, the opposite of what was intended. The competitive map is being redrawn in real time, and not obviously in the direction the policy hoped for.
The caveats matter, because the triumphant version of this story oversells it. First, "tops the download chart" is a measure of attention and availability, not of real-world dominance — a model can be the most downloaded thing on a hub while still trailing the best closed models on the hardest tasks, and the most eye-catching claims about these models come from their makers and their fans, not from neutral referees. Second, a model being free to download is not the same as it being usable. The largest of these systems are so big that running them at full strength requires a rack of expensive specialized chips almost no individual owns, the exact gap we described in the piece on open licenses and closed hardware. The hardware to run the best open models is itself subject to export controls. So the real picture is messier than "the ban backfired." Policy aimed at the software layer is leaking around the edges through open weights and a determined community, while a separate set of controls on the hardware layer still bites. The map is being redrawn — just not cleanly, and not yet in anyone's favor.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)