OpenAI has unveiled its next-generation GPT 5.6 household of AI fashions, however for maybe the primary time ever, the corporate’s most superior AI fashions will likely be out of attain for nearly everybody.
The brand new GPT 5.6 lineup consists of three mannequin variations, specifically: Sol, its flagship mannequin that’s stated to be essentially the most succesful; Terra, a mid-tier model that’s extra balanced for on a regular basis use; and Luna, a quick and reasonably priced model of the mannequin.
GPT 5.6 Sol is the best-performing mannequin out of all of them based mostly on testing on benchmarks associated to cybersecurity, biology, and agentic talents. It has additionally been developed with a “layered safeguard stack,” which appears to forestall dangerous actors from utilizing its AI mannequin for cyberattacks, amongst different malicious behaviours, based on OpenAI.
The general public launch of those fashions has been delayed indefinitely on the request of the US authorities, OpenAI stated in a weblog publish on Friday, June 26. Solely a small set of shoppers who’ve been pre-approved by the US authorities will have the ability to entry these fashions, as per the corporate.
OpenAI additional stated that it’s working with the Trump administration to regularly increase entry to a wider set of shoppers, together with worldwide companions, from subsequent week. The factors for the US authorities to approve GPT 5.6 clients is unclear, with OpenAI executives stating that it sends the federal government a listing and receives suggestions on it, based on a report by Wired.
OpenAI is the second frontier AI lab, after Anthropic, that has been compelled to restrict the supply of their most superior AI fashions on the behest of the US authorities. The incident has raised considerations that it might result in an unsure regulatory setting within the nation’s AI business, with consultants questioning how a lot energy a authorities ought to need to dictate AI mannequin releases.
Expressing displeasure with the request, OpenAI stated it hopes will probably be capable of make GPT-5.6 out there to everybody within the coming weeks.
Story continues beneath this advert
“We don’t consider this sort of authorities entry course of ought to turn into the long-term default. It retains the most effective instruments from customers, builders, enterprises, cyber defenders, and world companions who want them. We’re taking this short-term step as a result of we consider it’s the strongest path to broader availability within the coming weeks, whereas we work with the Administration to develop the cyber Government Order framework and a repeatable course of for future mannequin releases,” OpenAI stated.
Good new first: Sol is a brilliant, environment friendly, and a major step ahead. It’s the identical value as GPT-5.5. Additionally launching within the GPT-5.6 household is Terra, with 5.5-level efficiency at half the value.
Dangerous information: on the request of the US authorities, it’s launching right now in…
— Sam Altman (@sama) June 26, 2026
Describing the federal government entry course of as “not optimum”, OpenAI CEO Sam Altman stated, “We need to be a dependable, reliable associate that works with all stakeholders, and we additionally need to dwell by our mission of benefiting all of humanity. I consider the federal government shares most of our objectives, and that they’re general doing a very good job in a really tough scenario.”
What’s GPT 5.6 Sol?
OpenAI has named the three GPT-5.6 fashions after celestial our bodies: Sol, the Latin phrase for the Solar; Terra, the Latin title for Earth; and Luna, the Latin phrase for the Moon.
GPT 5.6 Sol comes with enhancements in agentic capabilities on coding, biology, and cybersecurity duties. It has two modes, a ‘max’ reasoning effort mode and an ‘extremely’ mode that makes use of coordinated sub-agents to resolve extremely complicated duties, based on OpenAI.
By way of efficiency, OpenAI excelled at a number of benchmarks. It additionally achieved scores that had been barely higher at coding workflows than Anthropic’s Claude Mythos 5, whose availability has additionally been restricted by the Trump administration. OpenAI additional claimed that GPT 5.6 Sol is on par with Mythos Preview whereas solely utilizing a 3rd of output tokens.
Story continues beneath this advert
On the protection entrance, OpenAI stated that GPT 5.6 Sol consists of its most strong safety stack but. It has been skilled in opposition to jailbreaking and adversarial assaults whereas being deliberately optimised to favour defensive cybersecurity work over offensive exploits. OpenAI additional stated that the protection guardrails have been constructed immediately into the core mannequin’s behaviour, somewhat than counting on a separate filter on high of it.
That is in distinction to Anthropic’s Claude Fable 5, whose security strategy boiled all the way down to mannequin classifiers detecting a high-risk subject corresponding to cybersecurity, biology, or chemistry, and re-routing the request to an older mannequin.
If or when the GPT-5.6 fashions are extra typically out there, OpenAI plans to offer entry to paid subscribers utilizing ChatGPT, Codex, and the OpenAI API. In a tiered pricing construction, OpenAI stated that Sol prices $5 per million enter tokens and $30 per million output tokens; Terra prices half that; and Luna prices $1 and $6, respectively. The corporate stated it has additionally improved immediate caching to make repeated prompts cheaper and extra predictable.
Whereas entry to the GPT 5.6 fashions is at the moment restricted, OpenAI stated it has rolled out a couple of updates to the default mannequin powering ChatGPT.
Story continues beneath this advert
Trump’s govt order and associated considerations
Earlier this month, US President Donald Trump signed an govt order that seemed to deal with the cybersecurity dangers posed by new and superior frontier AI fashions. The order acknowledged that the White Home would create a voluntary framework for AI corporations to share their fashions with the US authorities 30 days previous to their public launch.
Notably, the order emphasised that this assessment course of is fully voluntary and never a de facto licensing regime for AI mannequin releases. OpenAI and Anthropic had already been working with the U.S. authorities to check unreleased AI fashions, whereas Google DeepMind, Microsoft, and xAI (now SpaceXAI) agreed to offer the authorities early entry to new fashions for national-security evaluations.
Meta, then again, is the one main US-based AI mannequin supplier that has not reached an settlement to voluntarily share its fashions with the federal government for assessment, based on a report by The New York Instances.
In response to the US authorities stepping up oversight of the AI business, critics have argued that the dearth of clearly outlined security requirements might result in countless launch delays which may not solely give a hand to China within the AI race, but additionally jeopardise the billions of {dollars} being poured into AI infrastructure buildouts.