Prateek Tanmay
OpenAI has unveiled its next-generation GPT 5.6 family of AI models, but for perhaps the first time ever, the company’s most advanced AI models will be out of reach for almost everyone.
The new GPT 5.6 lineup includes three model versions, namely: Sol, its flagship model that is said to be the most capable; Terra, a mid-tier version that is more balanced for everyday use; and Luna, a fast and affordable version of the model.
GPT 5.6 Sol is the best-performing model out of all of them based on testing on benchmarks related to cybersecurity, biology, and agentic abilities. It has also been developed with a “layered safeguard stack,” which looks to prevent bad actors from using its AI model for cyberattacks, among other malicious behaviours, according to OpenAI.
The public release of these models has been delayed indefinitely at the request of the US government, OpenAI said in a blog post on Friday, June 26. Only a small set of customers who have been pre-approved by the US government will be able to access these models, as per the company.
OpenAI further said that it is working with the Trump administration to gradually expand access to a wider set of customers, including international partners, from next week. The criteria for the US government to approve GPT 5.6 customers is unclear, with OpenAI executives stating that it sends the government a list and receives feedback on it, according to a report by Wired.
OpenAI is the second frontier AI lab, after Anthropic, that has been forced to limit the availability of their most advanced AI models at the behest of the US government. The incident has raised concerns that it could lead to an uncertain regulatory environment in the country’s AI industry, with experts questioning how much power a government should have to dictate AI model releases.
Expressing displeasure with the request, OpenAI said it hopes it will be able to make GPT-5.6 available to everyone in the coming weeks.
Story continues below this ad
“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. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases,” OpenAI said.
Good new first: Sol is a smart, efficient, and a significant step forward. It is the same price as GPT-5.5. Also launching in the GPT-5.6 family is Terra, with 5.5-level performance at half the price.
Bad news: at the request of the US government, it is launching today in…
— Sam Altman (@sama) June 26, 2026
Describing the government access process as “not optimal”, OpenAI CEO Sam Altman said, “We want to be a reliable, dependable partner that works with all stakeholders, and we also want to live by our mission of benefiting all of humanity. I believe the government shares most of our goals, and that they are overall doing a good job in a very difficult situation.”
What is GPT 5.6 Sol?
OpenAI has named the three GPT-5.6 models after celestial bodies: Sol, the Latin word for the Sun; Terra, the Latin name for Earth; and Luna, the Latin word for the Moon.
GPT 5.6 Sol comes with improvements in agentic capabilities on coding, biology, and cybersecurity tasks. It has two modes, a ‘max’ reasoning effort mode and an ‘ultra’ mode that uses coordinated sub-agents to solve highly complex tasks, according to OpenAI.
In terms of performance, OpenAI excelled at several benchmarks. It also achieved scores that were slightly better at coding workflows than Anthropic’s Claude Mythos 5, whose availability has also been limited by the Trump administration. OpenAI further claimed that GPT 5.6 Sol is on par with Mythos Preview while only using a third of output tokens.
Story continues below this ad
On the safety front, OpenAI said that GPT 5.6 Sol includes its most robust security stack yet. It has been trained against jailbreaking and adversarial attacks while being intentionally optimised to favour defensive cybersecurity work over offensive exploits. OpenAI further said that the safety guardrails have been built directly into the core model’s behaviour, rather than relying on a separate filter on top of it.
This is in contrast to Anthropic’s Claude Fable 5, whose safety approach boiled down to model classifiers detecting a high-risk topic such as cybersecurity, biology, or chemistry, and re-routing the request to an older model.
If or when the GPT-5.6 models are more generally available, OpenAI plans to provide access to paid subscribers using ChatGPT, Codex, and the OpenAI API. In a tiered pricing structure, OpenAI said that Sol costs $5 per million input tokens and $30 per million output tokens; Terra costs half that; and Luna costs $1 and $6, respectively. The company said it has also improved prompt caching to make repeated prompts cheaper and more predictable.
