OpenAI confirmed late Tuesday that GPT-5.6 Sol, Terra, and Luna go generally available Thursday, July 9, ending a 12-day gated preview during which the models were released only to a small group of U.S.-based "trusted partners" whose names had been shared with the government. The Center for AI Standards and Innovation, housed inside the Department of Commerce, cleared the family after additional testing and a series of meetings with OpenAI technical staff who traveled to Washington. Axios first reported the clearance, citing a source familiar with the situation.

It's the first GPT launch conditioned on federal sign-off, and OpenAI's language about the process is careful. The company writes in its preview post that the arrangement "should not become the long-term default." Dean Ball, a former White House AI adviser, put it more plainly to TechCrunch: the setup functions as a "de facto involuntary licensing regime," activated by a Trump executive order asking frontier labs to voluntarily submit advanced models for review up to 30 days before release. The Anthropic precedent is instructive. Commerce briefly banned foreign access to Mythos and Fable in June, and lifted the restriction only last week.

The models themselves are aggressive. Sol claims a new state of the art on Terminal-Bench 2.1, a benchmark measuring command-line workflows that require planning and tool coordination, and introduces "ultra mode," which OpenAI describes as its first native multi-agent product, spawning subagents from within a single session. A new "max reasoning" effort setting is included. On Cerebras hardware, Sol runs at up to 750 tokens per second in July, capacity permitting.

Pricing tells the strategic story. Sol lists at $5 per million input tokens and $30 per million output. Terra sits at half those rates and is pitched as "competitive with GPT-5.5" at twice the price efficiency. Luna undercuts everything at $1 in and $6 out.

The Preparedness classifications are where the release genuinely breaks pattern. Per the system card: "Under our Preparedness Framework, we are treating Sol, Terra and Luna as High capability in both Cybersecurity and Biological and Chemical risk." A budget tier carrying the same risk designation as the flagship is new, and it reframes what "cheap model" now means in procurement conversations. None of the three cross the High threshold for AI Self-Improvement.

On ExploitBench, OpenAI says Sol is "competitive with Mythos Preview using only ~1/3 of the output tokens." Chain-of-thought controllability on ~5,000-token CoTs sits at 1.3% for Sol, against 0.4% for GPT-5.5 and 0.7% for GPT-5.4 Thinking; Terra remains at the lower controllability level of previous models. An updated system card is promised when the family goes broadly available.

The launch clears the review. The review's shape is now the story.

Sources

https://www.axios.com/2026/07/08/openai-gpt-trump-ban-lifted
https://openai.com/index/previewing-gpt-5-6-sol/
https://deploymentsafety.openai.com/gpt-5-6-preview
https://techcrunch.com/2026/06/26/openai-limits-gpt-5-6-rollout-after-government-request-says-restrictions-shouldnt-be-the-norm/
https://techxplore.com/news/2026-07-openai.html