OpenAI limits GPT-5.6 launch to US government-vetted users

TL;DR:

• OpenAI has begun a “limited preview” of its GPT-5.6 series — Sol, Terra and Luna — but only to a small group of partners approved by the US government.
• Chief executive Sam Altman called the staggered release “reasonable” but “not optimal”, objecting to the government choosing customers.
• Employees based in “supported countries” including the UK and Australia can use the model, even where their employers are US-based.

OpenAI has released its most advanced models yet to a tightly controlled list of users, at the request of the US government. The company announced GPT-5.6 Sol, its flagship, alongside the mid-tier Terra and lower-cost Luna, describing the rollout as a temporary preview while it works with Washington on a framework for future launches.

Government picks the customers

Only a small group of government-approved organisations has initial access through the API and OpenAI’s Codex tool, with all preview users based in the US. The Financial Times reported that commerce secretary Howard Lutnick intervened against even a limited release, pressing for sign-off from other agencies; the staggered approach followed talks with the Office of the National Cyber Director and the Office of Science and Technology Policy.

The move mirrors the treatment of Anthropic, whose Mythos model was pulled from general access under an export-control directive two weeks ago. Both followed President Trump’s executive order this month creating a voluntary framework to vet “covered frontier models” before release. OpenAI said it does not want this to “become the long-term default”, warning it keeps the best tools from developers, enterprises and cyber defenders.

Sol’s headline strength is cybersecurity. OpenAI says it shifts the performance-efficiency frontier on vulnerability research while staying below the firm’s “Cyber Critical” threshold, and is “better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks”. The company dedicated more than 700,000 GPU-hours to automated red-teaming for the release.

Looking forward

For UK businesses, the practical detail sits in the fine print: the Guardian reported that staff in supported countries, the UK among them, can access GPT-5.6 even when their employer is US-based, with broader international availability flagged for the coming weeks. Pricing starts at $5 per million input tokens for Sol. The episode underlines how frontier-model access is now gated by government clearance on both sides of the OpenAI–Anthropic rivalry — a constraint UK firms planning deployments will need to factor into their timelines.