OpenAI limits GPT-5.6 rollout after White House request, warns against permanent review regime

TechCrunch reports that OpenAI is holding the initial release of its GPT-5.6 lineup to a handful of trusted partners, a step it said on a Friday was taken at the US government's request. The lineup has three models: Sol, the flagship; Terra, a more balanced option for daily tasks; and Luna, quicker and cheaper. OpenAI said the preview is open only to partners whose involvement the government has been told about.

The article places the move within recent federal pressure on AI developers. After Anthropic released its Fable 5 model, the administration told the company to cut off foreign nationals, and Anthropic pulled the model entirely. Dean Ball, described as a former White House AI adviser about to join OpenAI, argues that President Trump's recent executive order, which invites certain companies to hand over advanced models for review a month before launch, has in practice produced an involuntary licensing regime. He warns that without clearly defined safety standards, launches could be delayed indefinitely, helping China and putting AI infrastructure investment at risk.

OpenAI went along but voiced displeasure in a blog post. It said such a government access process should not become the long-term default, since it denies the best tools to users, developers, businesses, cyber defenders and international partners. The company called the preview a short-term step toward wider availability within weeks, while it works with the administration on a cybersecurity executive order framework and a repeatable process for later releases.

The article also covers product details. OpenAI calls Sol its strongest model, with improved agentic skills in coding, biology and cybersecurity, and says its safeguards are built into the model's behavior itself. Sol's pricing is $5 for every million input tokens, with output tokens at $30 per million.