OpenAI is staggering the release of its GPT-5.6 artificial intelligence model by limiting early access to a small group of vetted partners approved by the US government, according to multiple sources [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]. The move came after the Trump administration requested a gradual and restricted launch to address national security risks associated with advanced AI capabilities [1, 2, 3, 4, 5, 6, 8, 9, 10, 11].
The GPT-5.6 suite consists of three models: Sol, the flagship version with the highest performance; Terra, positioned as a mid-tier balanced model; and Luna, optimized for speed and low cost [5, 6, 7, 8, 9, 10]. OpenAI carried out extensive safety testing on Sol, including approximately 700,000 A100e GPU hours of automated red-teaming, to enhance protections against malicious use [5, 6]. Sol is priced at $5 per million input tokens and $30 per million output tokens, significantly cheaper than competitor Anthropic’s Claude Fable 5 [5, 6].
Before the planned launch, OpenAI presented GPT-5.6’s capabilities and rollout strategy to US government agencies such as the Office of the National Cyber Director and the Office of Science and Technology Policy [2, 7, 8, 11]. In early June 2026, President Trump signed an executive order establishing a voluntary federal framework that requires AI models to undergo a 30-day government review before wider release [1, 2, 4, 8, 11].
OpenAI CEO Sam Altman said the phased release supported by the government is not the company’s preferred long-term approach. “We’ve made clear to the US government that this is not our preferred long term model, and will work with them and others in industry to achieve a more sustainable approach for future releases,” Altman said [1]. He also voiced reservations about the government deciding which customers get the model during the preview phase: “Extensive safety testing is not a bad idea. I just don’t like the idea of the government picking the customers” [2].
An official OpenAI blog cautioned against locking in government-mandated access controls, noting such restrictions block access for developers, businesses, cybersecurity professionals, and global partners who need advanced tools [6]. The company said it is taking this step short-term to enable broader availability soon while working on a repeatable federal review process under the forthcoming executive order framework [2]. A White House official confirmed ongoing collaboration with AI labs to develop common approaches to the challenges of scaling AI [11].
Anthropic, one of OpenAI’s competitors, also restricted access to its Mythos model under government pressure but fully withdrew it after foreign national access limitations were imposed [1, 2, 3, 4, 6, 8, 9, 11]. Separately, Meta reportedly faced government pressure to submit its AI models for voluntary security review [12].
The US Commerce Department’s AI Standards and Innovation Center is involved in reviewing and approving AI models before release [9, 10]. Meanwhile, OpenAI is bolstering its AI hardware efforts by hiring former Apple VP Paul Meade to lead AI-driven hardware alongside other ex-Apple executives [9, 10].
OpenAI plans to expand GPT-5.6 access more broadly in the coming weeks if the preview phase goes smoothly and a new government executive order framework is finalized [1, 2, 6, 7, 8, 9].
