Open-source AI is emerging as a beneficiary of the Trump administration’s recent rapid-fire policy changes towards AI regulation—and of rising costs. Developers and executives have told me that they’ve been burned by building apps on top of closed-source models like Anthropic's Fable 5, only to see the model withdrawn shortly after its release.

Developers are increasingly interested in using more open-source AI, since it’s hard for the government to prevent companies from using open-source models unless those companies do business with the government. It’s also cheaper: Coinbase CEO Brian Armstrong tweeted on Friday night that Coinbase was keeping its AI spending flat, despite rising usage, by experimenting with open models like Z.ai’s GLM 5.2 and Moonshot’s Kimi 2.7.

While leading open-source models are coming out of China, American open-source AI developers like Reflection AI, which is heavily backed by Nvidia, have been pushing the Trump administration to make less stringent regulations for open-source models because they’re not as advanced as closed-source AI, my colleagues and I have reported. (In this context, the Wall Street Journal report over the weekend that models from companies like Z.ai have matched Anthropic’s Mythos in some cybersecurity capabilities raises the stakes of the discussion.)