OpenAI ships GPT-5.6 Sol under a White House-gated preview, objects to the arrangement

The flagship, plus Terra and Luna, launched Friday under a government-managed access list — a first for a U.S. frontier model. OpenAI called the process an unsustainable default.

OpenAI launched GPT-5.6 Sol on Friday alongside two smaller siblings, Terra and Luna, behind a preview list vetted by the White House. It’s the first time a U.S. frontier model has shipped under a government-managed access arrangement, and OpenAI used its own announcement to object to the terms it was operating under.

“We don’t believe this kind of government access process should become the long-term default,” the company wrote, describing the setup as a short-term measure and promising broader availability “in the coming weeks.” The identities of preview participants, OpenAI acknowledged, had been shared with the administration. The partner list itself wasn’t published. Sam Altman told staff that access would be approved “customer by customer” during the preview, per The Information.

The specific asks came from the Office of the National Cyber Director and the Office of Science and Technology Policy, which according to CNBC and Axios wanted early-access limits while the administration builds out testing and evaluation procedures. The scaffolding is Trump’s June cyber executive order, which asks certain companies to voluntarily submit advanced models for government review up to 30 days before release. Dean Ball, a former White House AI adviser now joining OpenAI, has called the emerging structure a “de facto involuntary licensing regime.”

The parallel case is Anthropic, which this month pulled its Fable 5 model offline entirely after the administration ordered it to strip access from foreign nationals. Two labs, two different accommodations, one direction of travel.

On capability, OpenAI says Sol set a new state of the art on Terminal-Bench 2.1. Under its Preparedness Framework, all three models are rated High in Cybersecurity and Biological and Chemical risk, but below the Critical threshold; the system card notes Sol and Terra can locate vulnerabilities and partial exploits but couldn’t execute end-to-end attacks on hardened targets in testing.

Pricing, per TechCrunch: Sol runs $5 per million input tokens and $30 per million output; Terra sits at half those rates, $2.50 in and $15 out; Luna lands at $1 and $6. A Cerebras-hosted Sol running up to 750 tokens per second is due in July.

The tell isn’t the launch. It’s that the frontier lab publishing the model felt it necessary to append a disclaimer about the process required to publish it.

OpenAI ships GPT-5.6 Sol under a White House-gated preview, objects to the arrangement

Sources