OpenAI will release its GPT-5.6 model family to the public on Thursday, ending roughly six weeks of restricted access after the U.S. Commerce Department cleared the company for a broad rollout, according to Axios and OpenAI’s own confirmation.
The Commerce Department’s Center for AI Standards and Innovation carried out the additional evaluation that preceded the clearance, with OpenAI sending technical staff to Washington to respond to the agency’s questions, according to Axios. Commerce Secretary Howard Lutnick met separately with OpenAI chief executive Sam Altman to confirm that relevant federal agencies had reviewed and signed off on the model, the report said. Altman confirmed the release date in a post on X, writing simply, “GPT-5.6 Sol launches Thursday.”
GPT-5.6 launched in preview in late June under a naming structure new to OpenAI, three tiers named Sol, Terra and Luna rather than a single flagship model. Sol is positioned as the top-performing tier, built for complex coding, biology and cybersecurity tasks. Terra targets everyday enterprise workloads at lower cost, and Luna is aimed at high-volume, latency-sensitive tasks. Pricing is unchanged from the preview period: Sol costs $5 per million input tokens and $30 per million output tokens, Terra costs half that, and Luna costs $1 and $6.
Sol’s capabilities in coding, biology, and cybersecurity were among the factors that drew federal scrutiny in the first place, according to The Next Web. During the restricted preview, access was limited to roughly 20 partner organizations whose identities OpenAI shared individually with the government, an arrangement OpenAI has said was the first of its kind applied to an American frontier model. The review sits within a framework the Trump administration established June 2, calling for voluntary pre-release safety checks on the most capable AI systems; the GPT-5.6 case went further, moving from a voluntary check into a government-managed access list, which OpenAI agreed to only after being asked to delay the launch.
OpenAI has said it complied with the request but does not want the arrangement to become standard practice for future releases. “We don’t believe this kind of government access process should become the long-term default,” the company said in a blog post cited by TechCrunch. “It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”
The public release follows a comparable episode involving OpenAI’s chief rival. Anthropic suspended global access to its Claude Fable 5 and Mythos 5 models on June 12 under a separate U.S. export control directive, restoring access July 1 after implementing updated safety classifiers. OpenAI’s system for GPT-5.6 takes a different technical approach to the same underlying concern: rather than routing flagged high-risk prompts to an older, less capable model as Fable 5 does, OpenAI said Sol’s safety guardrails are built directly into the core model’s behavior rather than layered on as a separate filter, an approach the company has said is intended to reduce the false-positive routing issues that drew user criticism of Fable 5’s rollout.
Alongside the GPT-5.6 release, OpenAI also introduced a new generation of voice models called GPT-Live, which the company said can listen and speak simultaneously to create a more natural conversational experience. Sol will additionally become available on Cerebras hardware, where OpenAI expects inference speeds of up to 750 tokens per second for customers requiring higher throughput.
