Both of the Strongest Models Are Government-Gated
As the week ending June 29 wrapped up, a common thread surfaced: the strongest models from the two leading US labs are now both covered by Washington's new "frontier-model review" process. OpenAI previewed GPT-5.6 on June 26, with access government-gated; Anthropic's Mythos 5 was partially cleared on June 27, opened first to critical-infrastructure defenders. Neither model is currently open to the public.
GPT-5.6 Comes in Three Tiers
OpenAI split GPT-5.6 into three tiers this time: Sol is the flagship for cracking hard problems, Terra is the everyday balanced option, and Luna leads on speed, low cost, and high volume. Officially it's called the biggest architectural change since GPT-5. But note: right now it's a "preview," and access has to clear the government gate—not everyone can use it.
Mythos 5 Partially Cleared
On Anthropic's side, Mythos 5 was forced offline on June 12 over export controls, and after two weeks of back-and-forth with the Commerce Department, obtained partial clearance on June 27, for use first by critical-infrastructure defenders (per NPR reporting). The lawsuit with the Defense Department over the model's use is still proceeding on a separate legal track. Putting the two together, the real change this week isn't how strong the models are, but that "who can use them" is starting to be gated by government. For practitioners, this means the availability of frontier models is increasingly governed by export controls and safety review—so when selecting, besides capability and price, you have to factor in "can I get it, and is it compliant." Specific developments are per the formal disclosures of OpenAI, Anthropic, and US officials.
via: NPR