OpenAI introduced Friday that it’ll restrict the discharge of its newest AI fashions to a “small group of trusted companions” on the request of the U.S. authorities.
The subsequent technology GPT-5.6 lineup contains the flagship mannequin Sol. Terra, a extra balanced mannequin for on a regular basis use. Luna is a sooner, decrease value choice. The Sol is the corporate’s strongest mannequin, however the Trump administration has restricted gross sales of all three fashions. OpenAI mentioned the preview can be restricted to companions that “share participation with governments.”
The administration’s request comes amid renewed stress from the U.S. authorities to restrict cutting-edge methods on AI corporations. After Anthropic launched its strongest public mannequin, Fable 5, the federal government ordered the corporate to take away international entry, main Anthropic to take away the mannequin fully.
The incident raised questions on how a lot energy the federal government ought to have over the discharge of AI fashions. Dean Ball, a former White Home AI advisor and soon-to-be OpenAI worker, says President Trump’s current government order requiring some AI corporations to voluntarily submit their newest fashions for presidency evaluation 30 days earlier than launch creates a de facto involuntary licensing system for frontier AI, resulting in heavy-handed restrictions.
Ball argues that the issue can be exacerbated if governments would not have clearly outlined security requirements, which may result in countless launch delays, handing China a hand within the AI race in addition to jeopardizing billions of {dollars} invested in AI infrastructure.
Now, OpenAI has complied with the administration’s calls for, however the AI firm has made it clear that it’s not pleased with the deal.
“We don’t consider this sort of authorities entry course of ought to grow to be the long-term default,” Friday’s weblog publish mentioned. “We safe the most effective instruments from the customers, builders, enterprises, cyber defenders, and world companions who want them.”
OpenAI mentioned the corporate is working with the administration to develop a brand new government order framework for cybersecurity and a “repeatable course of for future mannequin releases,” and known as this preview a “short-term step” that can put GPT-5.6 on the trail to broader availability within the coming weeks.
GPT-5.6 Sol specs
OpenAI says GPT-5.6 Sol is its strongest mannequin up to now, with improved agent capabilities in coding, biology, and cybersecurity. Sol introduces a “most” inference effort mode and an “extremely” mode that makes use of tuned subagents to resolve extraordinarily advanced duties (a sort of neat trick that spikes token utilization).
OpenAI says GPT-5.6 outperforms a number of benchmarks, together with being barely higher at coding workflows than Anthropic’s Claude Mythos 5, which was additionally successfully banned by the Trump administration this month. GPT-5.6 Sol additionally competes with Mythos Preview, however makes use of one-third of the output tokens, in line with OpenAI.
To allay considerations that the highly effective mannequin is insecure, OpenAI says Sol contains its most sturdy safety stack up to now. OpenAI says it’s considerably hardened in opposition to adversarial assaults and deliberately optimized to prioritize defensive cybersecurity efforts over offensive exploits. In different phrases, it’s designed to be tough to jailbreak whereas prioritizing displaying customers find out how to shield in opposition to exploits quite than find out how to hack into their methods.
OpenAI additionally says its security guardrails are constructed immediately into the core mannequin’s habits, quite than counting on a separate filter on high of the core mannequin. The corporate is probably going attempting to keep away from the entice that caught Anthropic with Fable 5. Within the temporary second that Fable 5 was obtainable, when the mannequin’s classifier detected a high-risk subject like cybersecurity, biology, or chemistry, it will merely block the immediate; Requests can be routed to the previous mannequin. The general over-cautious circulation and invisible downrouting resulted in lots of false positives and person backlash.
The GPT-5.6 mannequin will initially be obtainable solely to a choose group of companions, however OpenAI plans to make it extra broadly obtainable to these utilizing ChatGPT, Codex, and APIs within the close to future.
GPT-5.6 is on the market in three sizes with tiered pricing. Sol prices $5 per million enter tokens and $30 per million output tokens. Terra prices half that. Luna costs are $1 and $6, respectively. OpenAI says it has additionally improved immediate caching, making repeated prompts cheaper and extra predictable.
If you happen to purchase by means of hyperlinks in our articles, we might earn a small fee. This doesn’t have an effect on editorial independence.
