My jaw dropped when I read of Z.ai’s GLM-5.2 model performance. The model by Beijing-based Zhipu AI was released on June 13, and since then I’ve been reading report after report of its real world performance. It is an open-weight model and can be used online super cheap (or free if you just sign into their website). It can also be downloaded and run on local hardware, free. The performance is incredible, and that is reflected in benchmarks. It is the top open-weight model on coding and long-horizon agentic benchmarks. It is just barely behind the best closed models (Fable 5, Opus 4.8, GPT-5.5).

Want to do serious work via API? You can do that for roughly one tenth the cost of Anthropic or OpenAI’s best models.

So, incredible performance, incredibly low cost.

And it is Chinese, not American in any way.

Many American business leaders have already concluded they are spending too much on Anthropic and OpenAI and have already switched to open weight models. Others are watching closely and will likely do so in the near term.

Coinbase CEO Brian Armstrong recently described how he is shifting his architecture for developers to let them choose models to keep his AI spend flat. He did not announce a complete shift to GLM 5.2, but it is definitely part of his developer tooling. How can you blame him or any other CEO for making decisions like this?

Z.ai’s GLM is just one of many PRC models. Chinese labs release more open-weight models than the rest of the world combined. Eight Chinese labs, including DeepSeek, Alibaba’s Qwen, Moonshot’s Kimi, Xiaomi’s Mimo, and Zhipu, have collectively released more MIT-licensed and Apache 2.0-licensed open-weight models in 2026 than all non-Chinese labs combined. Six of the models now appear on major AI capability rankings.

With all these incredible open-weights capabilities out there, why should any U.S. business be forced to select from frontier model companies with closed weights offerings? Is there some business benefit in doing that? I hope I don’t sound like a socialist there, I am definitely a supporter of free enterprise capitalism. But that includes supporting capitalists who are building elsewhere in the economy. Think of every industry of the economy. Should they all have to pay Anthropic or OpenAI millions to do what they can do for a tenth the cost?

I’m also pro-America and do not want the PRC to win this race, but if we don’t rethink our approaches it looks like it could be game over. The U.S. approach has shunned open source, with Llama being the only open weights alternative for enterprises. And the fact is that Llama’s best model does not come close to GLM 5.2 performance.

So what can the nation do about this? Do we surrender? No.

In my view the nation should shift to a policy of supporting open source, open weights model development. This was the master move of the PRC. Many of us had been advocating for this for years. The global economy has now spoken. It is the right path.

There are many other open questions. What does this mean for the business models of OpenAI and Anthropic? They are already subsidizing token costs as they try to close new business. And businesses are already realizing they have to cut those costs. Even without the PRC open weights models OpenAI and Anthropic are facing business pressure. Now that those models are so cheap what will that do to the equation? What will this mean for their IPOs?

And what does this mean for security researchers? Shouldn’t they all have the power they need from America’s best models? We have all been watching the ongoing chaos over how Anthropic and OpenAI models can find vulnerabilities in all software. And watched the government’s actions with Anthropic (and later OpenAI) to try to keep these capabilities from getting into the hands of the PRC. Now this makes that entire argument moot. What are we trying to keep them from getting? They have it already.

The U.S. has spent years protecting a closed-model approach that the PRC just rendered obsolete. The window to course-correct is open, but not indefinitely. A national commitment to open-weight model development, funded, coordinated, and treated with the same urgency as the semiconductor supply chain, is the master move. Everything else is delay.