OpenAI's newest models are now open to everyone — and the story of how they got there says as much about AI's new reality as the models themselves.
All three GPT-5.6 variants — Sol, Terra and Luna — went live for ChatGPT users and API developers on Thursday, July 9. "We're expanding preview access globally now," the company posted on X. It caps off an unusual staggered launch: in late June, the models had reached only what OpenAI called a "small group of trusted partners."
Why the staggered launch?
That slow rollout wasn't marketing. It goes back to an AI cybersecurity order President Trump signed in early June, asking companies to voluntarily hand their most powerful models to the government for review 30 days before any public release.
So GPT-5.6 initially went only to select government-approved entities. OpenAI didn't hide its discomfort. "We don't believe this kind of government access process should become the long-term default," the company said at the time — but it played along, betting that cooperation was the quickest path to a full launch.
The bet paid off early. According to Axios, the administration cleared the wider release in under 30 days, after the Commerce Department's Center for AI Standards and Innovation ran extra tests on the models. OpenAI reportedly flew technical experts to Washington so questions got answered on the spot instead of over weeks of back-and-forth.
Three models, three jobs
GPT-5.6 isn't one model. It's a family, and each member has a job.
Sol is the flagship — OpenAI's strongest model yet. It's priced like it, too: $5 per million input tokens, $30 per million output.
Terra is the workhorse. OpenAI says it matches GPT-5.5-level performance at half the cost — $2.50 per million input tokens, $15 per million output.
Luna is the budget pick for high-volume, lighter tasks: $1 in, $6 out.
Look at those numbers for a second. By selling Terra as a GPT-5.5-class model at half the price, OpenAI is squeezing the market from the middle — capable AI gets cheaper for most apps, while Sol handles the genuinely hard problems.
The new normal for AI launches
GPT-5.6's route to the public also previews how frontier AI releases may work from here on. OpenAI isn't alone in dealing with Washington. Anthropic had to block all access to its latest Mythos cybersecurity and Fable models when the government ordered it to keep foreign nationals out. It has since won permission to redeploy the more exclusive Mythos 5, with Fable 5 — the version that brings many of those capabilities to the public — to follow.
Will this review process become permanent? Nobody knows, and the labs are visibly uneasy about it. What's clear is that shipping a major AI model is no longer purely a company decision. There's a stop in Washington now.
What it means for users
If you use ChatGPT, the change just shows up — the new models are simply there. Developers face a more interesting choice: Terra's price-performance ratio makes it the obvious default for most applications, with Sol saved for frontier reasoning and long-horizon work.
For everyone else? The world's most advanced AI models now ship with a government review stamp. And this time, at least, the process moved faster than anyone expected.
Image: via Pexels
