OpenAI has opened GPT-5.6, the most advanced model family it has commercially offered to date, to wide release. The family consists of three models: the most powerful version, Sol; a mid-tier, more affordable option called Terra; and the most economical variant, Luna.
The model was introduced at the end of June, but access was initially shared only with a limited number of approved partners. The reason was a review process in which the US government assessed the risk of increasingly powerful AI systems being misused for cybersecurity or military purposes. A similar process had previously occurred with the launch of Anthropic's Fable model.
Why does it matter?
This is a sign of a new pattern in how increasingly powerful AI models are being released. Through an executive order signed in June, the US administration established a voluntary framework requiring developers to share "covered frontier models" with the government for a certain period before wide public release.
From a business perspective, this kind of review can make model release timelines unpredictable — which once again shows that "switching to the newest model" isn't, on its own, a reliable strategy.