OpenAI is releasing a new AI system called GPT-5.6 on Thursday after getting approval from the U.S. government. The company had kept it available only to a small group of trusted users for two weeks while government officials reviewed it.

OpenAI is actually releasing three versions of GPT-5.6, each designed for different needs and priced differently. Sol is the most powerful and most expensive version, aimed at professional developers and large companies. Terra offers solid performance at a lower price, targeting everyday users and businesses. Luna is the cheapest option for budget-conscious users.

Why the delay? In early June, the Trump administration asked major AI companies to let the government review their most powerful models 30 days before making them public. OpenAI followed this request, which is why GPT-5.6 was only available to a small group while it went through government testing.

The review happened at the U.S. Department of Commerce's Center for AI Standards and Innovation. Government officials tested the model and asked OpenAI questions about how it works and how safe it is. OpenAI sent engineers to Washington to answer those questions in person. Eventually, the government approved the release.

Other companies have gone through this same process. Anthropic, another major AI company, had similar restrictions placed on two of its models. The government asked Anthropic to block certain users from accessing them. After addressing those concerns, Anthropic was eventually allowed to release the models more widely.

This is a new way of doing things. In the past, AI companies made their own decisions about when and how to release new models. They did not need government approval before launching. Now, there is a 30-day waiting period where the government can review powerful AI systems before the public gets access.

This change raises questions about what happens next. Will this become the normal way AI models are approved going forward? Or is this just a temporary rule tied to the current president's policies? OpenAI has said that having these kinds of restrictions "shouldn't be the norm," which suggests the company is not happy with this process. But they are still following it.

For companies buying and using these models, the pricing matters most right now. Luna and Terra are much cheaper than Sol, so many organizations will probably use those for everyday work. They will only use the more expensive Sol model for specific tasks where its extra power really matters.