OpenAI has introduced its next-generation GPT-5.6 artificial intelligence models, but the initial rollout will be limited to a select group of US-based partners after consultations with the US government.

The GPT-5.6 family includes Sol, the company’s flagship model, Terra, a mid-tier model for general-purpose tasks, and Luna, a lower-cost, high-speed option. OpenAI said Terra will be offered at half the price of GPT-5.5 as it looks to strengthen its position against rivals such as Anthropic and Google.

According to the company, the new models were reviewed with US officials before launch. At the government’s request, OpenAI is starting with a restricted preview available only to trusted US organizations whose identities have been shared with authorities. Overseas employees of those organizations will also be able to access the models.

The limited release follows recent US action against Anthropic, which was instructed to prevent foreign nationals from using its latest AI models over national security concerns.

Advanced AI systems from both OpenAI and Anthropic have reportedly demonstrated an unprecedented ability to identify software vulnerabilities, prompting increased government scrutiny over the potential cybersecurity risks posed by cutting-edge AI technologies.