OpenAI restricts GPT-5.6 rollout to trusted partners

The government-coordinated restriction that made this story notable lasted about two weeks: OpenAI took GPT-5.6 Sol, Terra, and Luna to full general availability on July 9, 2026, which is the more consequential data point for practitioners than the initial access controls, since it sets the actual timeline other frontier labs can be measured against for any future pre-release government review process.

OpenAI announced three new models, GPT-5.6 Sol, Terra, and Luna, on June 26, 2026, and said it would initially limit access to a "small group of trusted partners," per CNBC. CNBC reports OpenAI previewed the models' capabilities and shared its plans with the U.S. government ahead of the public launch. Politico reports the company changed course after a White House request, made in consultation with the Office of Science and Technology Policy and the Office of the National Cyber Director; Bloomberg similarly reports U.S. officials asked OpenAI to stagger the release. CNBC quotes OpenAI: "We don't believe this kind of government access process should become the long-term default."

Per OpenAI's own help-center documentation and independent corroboration (Simon Willison's analysis), the preview - initially limited to roughly 20 vetted organizations - ended on July 9, 2026, when OpenAI made all three tiers generally available across ChatGPT, Codex, and the API, with global rollout completing within 24 hours. Per-million-token GA pricing is $5/$30 for Sol, $2.50/$15 for Terra, and $1/$6 for Luna, and OpenAI added more predictable prompt caching with explicit cache breakpoints and a 30-minute minimum cache life.

Multiple outlets framed the original restriction as a continuation of heightened U.S. government scrutiny of frontier models, alongside Anthropic's recent decision to disable access to two of its newest models under an export-control directive. Together, those facts indicated regulators and major model developers were testing mechanisms for pre-release capability review and controlled rollouts. The roughly two-week duration of OpenAI's restriction - rather than an open-ended gate - suggests this specific review mechanism functioned as a delay rather than a lasting access limitation.

Teams that catalogued dependency risk or shifted to alternative models during the preview window can now access GPT-5.6 Sol, Terra, and Luna directly through the API at the published GA pricing; the roughly two-week gap between announcement and full availability is a useful reference point for estimating how long a similar government-coordinated review might take for a future frontier release.

Whether the White House or Office of Science and Technology Policy ever publishes formal, public criteria for its capability-assessment process, since none of the cited reporting indicates one exists; and whether other frontier labs face comparable pre-release reviews with similarly short turnaround times.