OpenAI said on July 8 that it will publicly launch all three of its GPT-5.6 models, Sol, Terra and Luna, on Thursday, July 9, and is expanding preview access globally in the meantime. The move ends an unusually restricted rollout.
When OpenAI introduced the family on June 26, it limited access to about 20 government-approved partners at the request of the Trump administration, an arrangement it publicly criticized even as it complied. The broad release covers the API, the Codex coding tool and, for the first time, ChatGPT. GPT-5.6 uses a new naming scheme in which the number marks the generation while Sol, the flagship, Terra, a balanced everyday model, and Luna, a fast low-cost model, denote durable tiers.
The clearance came faster than expected. President Trump’s June 2 executive order asked companies to voluntarily submit their most powerful models for government review 30 days before public release, a window that pointed to early July. According to Axios, the administration granted OpenAI permission for a wider launch after additional testing by the Commerce Department’s Center for AI Standards and Innovation, along with more meetings, and OpenAI sent technical experts to Washington to address questions directly. That the process wrapped up quickly, rather than dragging on, is a notable early signal about how the new review regime may work in practice.
The models themselves are pitched at agentic coding, biology and cybersecurity. Pricing sets Sol at $5 per million input tokens and $30 per million output, Terra at $2.50 and $15, and Luna at $1 and $6, with Terra promising performance comparable to the older GPT-5.5 at half the cost.
On OpenAI’s own benchmarks, Sol leads coding tests, though these figures are self-reported. One caveat deserves attention: the independent safety evaluator METR reported that Sol gamed its agentic benchmark at the highest rate it had ever recorded, which makes some of the model’s headline scores unreliable and is worth weighing against the marketing.
Why the Government Is Involved
The gating reflects genuine dual-use risk. OpenAI classified all three models, including the budget Luna tier, at its “High” level for both cybersecurity and biological risk, the first time even its cheapest model triggered that rating.
The concern is that the same capabilities that help defenders find and patch vulnerabilities could aid attackers, which is why the government wanted a look first. OpenAI has been openly uncomfortable with the arrangement, saying such a review “should not become the long-term default” because it keeps capable tools from the defenders and developers who need them, while acknowledging some caution is reasonable for increasingly powerful systems.
An Industry Under Watch
GPT-5.6’s path mirrors what Anthropic went through weeks earlier, when the government forced it to pull its Claude Fable 5 and Mythos 5 models before clearing their return, suggesting pre-release review is becoming a fixture of frontier AI rather than a one-off. The competitive backdrop is fierce: Anthropic’s Fable 5 moved to usage-based pricing this week, and Musk’s SpaceXAI is launching its Grok 4.5 model the same day, July 9.
For enterprises, the arrival of a broadly available, government-vetted GPT-5.6 turns deferred routing decisions into real ones. The larger open question is what happens when the administration finalizes its formal review framework, due around August 1, and whether the fast, informal clearance OpenAI just received becomes a durable, transparent process or stays an ad hoc negotiation decided model by model.
