GPT-5.6 Was Locked Behind a Government Gate. It's Open Now

9 min read By FindSkill Team

GPT-5.6 is public as of July 9 — the government review is over. What the gate was, who really decided, and what to check if you still can't see it.

Table of Contents

Update — July 10, 2026: the gate is open. GPT-5.6 went public on July 9 — in ChatGPT, Codex, and the API, rolling out globally over 24 hours. If you searched “why can’t I use GPT-5.6,” the answer changed this week: you can. The quick access check is below, and the rest of this page is the story of the strangest AI launch to date — kept here because it’s about to become the template for every big model release.

The short answer, July edition

You can use GPT-5.6 now. On July 9, OpenAI released Sol, Terra and Luna publicly, ending a thirteen-day stretch in which the most capable model on the market was limited to roughly 20 pre-approved organizations while a new U.S. government review process ran its course.

If it’s still not showing up for you, it’s almost certainly one of these three, in order of likelihood:

  1. The rollout reached you late. It was staged over 24 hours from July 9, and some users saw the new models appear in ChatGPT Work or Codex hours before regular chat.
  2. You’re looking in the wrong place for your plan. In regular chat, GPT-5.6 Sol is for Plus, Pro, Business and Enterprise. On Free or Go, you won’t see 5.6 in normal chat at all — your access is through ChatGPT Work or Codex, where free users get Terra. That’s not a bug; that’s the mapping.
  3. You’re expecting a “5.6” toggle that looks like the old picker. The lineup is now Sol / Terra / Luna plus effort settings. Our Sol vs Terra vs Luna guide covers which to pick.

That’s everything a person searching this headline needs. What follows is the part worth keeping: what actually happened here, because it had never happened before and it will happen again.

What the gate actually was

On June 26, OpenAI launched GPT-5.6 as a limited preview to about 20 organizations, through the API and Codex only — not ChatGPT. No waitlist. No paid tier that unlocked it. Even well-known AI companies publicly noted they weren’t on the list.

GPT-5.6 access: the full arc
June 26 — gated preview ~20 approved orgs · API + Codex only
Government review Commerce Dept's AI standards center · extra testing
July 9 — public release ChatGPT, Codex, API · global in 24 hours
June 26 to July 9 — thirteen days from gated preview to public release.

The backdrop was an executive order signed June 2, 2026, directing federal agencies to build an evaluation process for powerful new AI models — with companies asked to submit frontier models for review up to 30 days before public release. An earlier draft had said 90 days; the final order said 30. GPT-5.6 was the first major model to walk through that door.

During the review, the Commerce Department’s Center for AI Standards and Innovation ran additional testing, and OpenAI sent technical staff to Washington for meetings. Thirteen days after the gated launch, the public release followed.

Who actually decided — the part everyone got wrong

Here’s the detail that makes this episode more interesting than “government approves AI model.”

When the July 9 release was announced, reporting described the administration as having given OpenAI permission for the wider rollout. The White House pushed back on that framing, hard — stating it did not give OpenAI a “green light,” that “no such permission is required or granted,” and that release decisions “rest entirely with the companies.”

Both things are technically true at once, which is exactly why it’s confusing:

  • The executive order’s framework is formally voluntary. It explicitly does not create a licensing or clearance requirement. Legally, OpenAI could have shipped publicly on day one.
  • In practice, OpenAI chose to hold the public release until the review concluded — calling it “the strongest path to broader availability” while also saying, pointedly, that it doesn’t believe “this kind of government access process should become the long-term default.”

So: not a legal gate, but a very real one. A voluntary process that a company treats as unskippable is a new kind of thing in software launches, and the vocabulary for it doesn’t exist yet. “The government approved it” is wrong. “The government had nothing to do with the timing” is also wrong. The honest sentence is: OpenAI waited until the government’s voluntary review finished, and both sides are keen you don’t call that permission.

Why this stays relevant after the gate opened

Because it’s the template now. The 30-day framework didn’t expire when GPT-5.6 shipped — it’s the standing process for frontier models. When the next big release from OpenAI, Anthropic or Google seems oddly quiet for a few weeks after being announced, this is likely why. You’ll recognize the shape: limited preview, “trusted partners,” a review window, then the real launch.

Worth watching over the next releases:

  • Does “voluntary” stay voluntary? The White House’s insistence that no permission is required is doing a lot of work. If a future review runs long and a company ships anyway, we learn what the framework really is.
  • Does the two-tier window shrink or grow? Thirteen days was short enough that most people barely noticed. A 30-day-maximum window used in full, every release, is a different experience.
  • Do other governments copy it. A model gated in the U.S. while shipping openly elsewhere would restart this whole debate with the countries swapped.

What this episode didn’t mean

  • It didn’t mean ChatGPT was being restricted. Your account worked normally on GPT-5.5 throughout.
  • It didn’t mean insiders were quietly using 5.6 for months. The preview group was small, named to the government, and access ran through developer tools.
  • It didn’t mean the government wrote or changed the model. The review evaluated capabilities — particularly around cybersecurity — it didn’t edit the product.
  • It didn’t set a legal precedent. It set a behavioral one, which may matter more.

The bottom line

The question this page was built on — “why can’t I use GPT-5.6?” — died on July 9. The answer that replaced it is worth the two minutes: the first frontier model shipped under Washington’s voluntary review framework, the gate held for thirteen days, and both OpenAI and the White House came away insisting nobody needed anyone’s permission while acting exactly as if they did.

GPT-5.6 itself is now just… available. What Sol, Terra and Luna each do, what’s new in the release, and the ChatGPT Work agent it shipped with are the useful next reads.

And the advice from the waiting period aged fine: the people who spent those two weeks getting better at the tool they had are the ones getting the most out of the upgrade this week. AI Fundamentals and ChatGPT vs Claude still make you better on whatever model you’re on.

Sources

Build Real AI Skills

Step-by-step courses with quizzes and certificates for your resume