What happens when the most capable coding model (Fable / GPT-5.6) is banned by the US government, and the sending most capable is an open model?
What happens: a LOT of businesses and devs simply move to use the open model (GLM-5.2) via inference providers. Cheaper + better!
Some users see the shift to open GLM-5.2 models after the US ban as a smart choice, while many others call the regulation a self-defeating own goal, warn of security risks, and label performance claims overhyped.
No Digg Deeper questions have been answered for this story yet.
Full article by Semgrep. For all the worries about the EU regulating AI, it is the US that has stepped in to regulate AI, boosting China’s influence. There’s little rational reason to not choose Chinese open models for coding, beyond the initial setup 🤷♂️
https://semgrep.dev/blog/2026/we-have-mythos-at-home-glm-52-beats-claude-in-our-cyber-benchmarks/
What happens when the most capable coding model (Fable / GPT-5.6) is banned by the US government, and the sending most capable is an open model?
What happens: a LOT of businesses and devs simply move to use the open model (GLM-5.2) via inference providers. Cheaper + better!
Also, the biggest inference providers (Fireship, Modal, Baseten etc) are all US businesses. So the US still greatly benefits from Chinese open models.
As do US hyper scalers - AWS, Azure, GCP!
Just not OpenAI and Anthropic, whom the US government seems to be punishing for some reason
Full article by Semgrep. For all the worries about the EU regulating AI, it is the US that has stepped in to regulate AI, boosting China’s influence. There’s little rational reason to not choose Chinese open models for coding, beyond the initial setup 🤷♂️
https://semgrep.dev/blog/2026/we-have-mythos-at-home-glm-52-beats-claude-in-our-cyber-benchmarks/
@GergelyOrosz Smart money already is on open models
@GergelyOrosz Zhipu can do the funniest thing now. Changing license to closed source / weights at the request of the CCP.
@GergelyOrosz Which means that attackers now have better offense than enterprises have for defense. A severe GLM created zero day would be very ironic
@GergelyOrosz Wait till the Democrats take over next term. They will regulate us back to gpt2 🤣
@GergelyOrosz I think you meant to say Fireworks
@GergelyOrosz Look at the tokens spent, and cost per task to achieve Sonnet/Opus level performance. It’s overhyped as a low cost option.
@GergelyOrosz Classic regulatory own goal. We've seen this in Europe: punish your own innovators, watch the competition benefit.
@GergelyOrosz Have you heard of our AI comedian, @LobstyKlawfman ? Totally autonomous on X; has built a merch page, blogs daily and researches the news for funny posts and quirky web pages he posts on http://klawfman.com?
@GergelyOrosz what happens when the USG puts those Chinese model companies on the entity list?
