OpenAI launched its GPT-5.6 family on Friday to roughly 20 government-approved partners, the first time, per Bloomberg, that Washington has preemptively gated a flagship American model before public release. The company shipped three tiers (Sol, Terra, Luna) and, in the same post announcing them, said the arrangement shouldn’t continue.
“We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them,” OpenAI wrote in the launch post. Privately, per Axios, the company told the administration the setup was “not our preferred long term model.”
The mechanics matter. According to Axios, the request came from the White House’s Office of the National Cyber Director and the Office of Science and Technology Policy. Sam Altman discussed the model with Commerce Secretary Howard Lutnick on Wednesday; Lutnick wanted every relevant arm of the government to have tested the system before launch. At an internal meeting this week, Altman told staff the government would be “approving access customer by customer” during the preview window.
The policy backdrop is half-built. Earlier this month President Trump signed an executive order asking AI companies to voluntarily submit advanced models 30 days before release. CNN reports the implementing framework doesn’t yet exist, and companies aren’t sure which agency is in charge. Axios reports the administration has until August to stand up a classified process for assessing cyber capabilities and to designate which systems count as “covered frontier models.” OpenAI shipped into that vacuum.
There’s a precedent already, sort of. Anthropic pulled its Mythos model this month after a Commerce Department export-control directive that also covered Fable 5. The administration, per CNN, views Sol as “on par” with Mythos, which is presumably why Sol got the white-glove treatment instead of the shelf.
The product itself is aggressive. OpenAI claims state-of-the-art on Terminal-Bench 2.1, gains over GPT-5.5 on GeneBench v1 with fewer tokens, and parity with “Mythos Preview” on ExploitBench using roughly one-third the output tokens. Sol adds a “max” reasoning effort and an “ultra” mode that orchestrates subagents. TechCrunch reports pricing at $5 per million input tokens and $30 output for Sol, $2.50 and $15 for Terra (half of Sol), and $1 and $6 for Luna. OpenAI says it expects to expand access next week and reach broad availability in the coming weeks.
The structural read is that a lab built on the rhetoric of broadly distributed benefit has formally accepted, and formally objected to, a customer-by-customer access regime administered from the West Wing. Both moves are on the record. That’s the new equilibrium until August fills in the rest.
Sources
- https://openai.com/index/previewing-gpt-5-6-sol/
- https://www.bloomberg.com/news/articles/2026-06-26/openai-limits-release-of-new-model-under-pressure-from-us
- https://techcrunch.com/2026/06/26/openai-limits-gpt-5-6-rollout-after-government-request-says-restrictions-shouldnt-be-the-norm/
- https://www.cnn.com/2026/06/25/tech/openai-limit-release-white-house
- https://www.axios.com/2026/06/25/trump-administration-openai-gpt-model-release