OpenAI released its most advanced model series to the public today: GPT-5.6, including its flagship Sol model and the lower-tier Terra and Luna models. The launch comes weeks later than planned — and the delay had an unusual sender.
Why it was delayed
The Trump administration asked OpenAI to keep GPT-5.6 limited to a small set of government-approved partners before any wider release, citing security concerns. It was the first time the U.S. government preemptively asked an American AI company to throttle a model launch.
The reason: both OpenAI and the administration view GPT-5.6 as ‘on par’ with Anthropic’s Mythos — especially in cybersecurity. Those very capabilities triggered the request for a controlled, staggered rollout. The model was initially confined to roughly 20 individually cleared partners. Testing was run by the Center for AI Standards and Innovation inside the U.S. Department of Commerce.
The timing is no accident
OpenAI had internally locked its target window between July 7 and July 9. July 7 was the first day Anthropic’s Claude Fable 5 dropped out of subscription bundles and switched to pay-as-you-go. In other words: OpenAI hits the button exactly when Anthropic’s top model gets more expensive for a lot of users.
My take
Two things stand out here. First: a government asking an AI company to hold back a model — and a company that plays along but insists this shouldn’t become the norm. That’s a new level of entanglement between AI and the state, and we’ll probably see more scenes like it.
Second: the timing against Fable 5. The model race isn’t decided by benchmarks alone anymore — it’s price, availability, and picking the right moment. If you have a strong model but launch at the wrong time, you leave attention on the table. OpenAI clearly gets that.
Sources: