OpenAI has announced that it is limiting access to its upcoming GPT-5.6 AI models after a request from the U.S. government, allowing only a small group of trusted partners to test the new systems before their wider release.
The GPT-5.6 family includes three models: Sol, the company’s most advanced flagship model, Terra, which is designed for balanced everyday use, and Luna, a faster and more affordable option. Although Sol is the most capable model in the lineup, all three versions are currently restricted to selected partners whose participation has been shared with the U.S. government.
The move follows increasing pressure from the Trump administration on AI companies to tighten control over their most powerful models. Earlier this month, Anthropic was forced to remove public access to its Fable 5 model after the government ordered the company to restrict availability to foreign nationals. That decision sparked debate over how much influence governments should have over the release of frontier AI technologies.
Dean Ball, a former White House AI adviser who is set to join OpenAI, argued that a recent executive order asking AI developers to voluntarily submit advanced models for government review up to 30 days before launch has effectively created an unofficial licensing system. According to Ball, the lack of clearly defined safety standards could delay future AI releases, slow innovation, and potentially give China an advantage in the global AI race while also affecting billions of dollars in AI infrastructure investments.
OpenAI said it complied with the government’s request but made it clear that it does not want this process to become permanent. The company stated that limiting access prevents developers, businesses, cybersecurity professionals, and global partners from using its latest technology when they need it most.
The company described the restricted preview as a temporary measure and said it is working with the administration on a new cybersecurity framework and a more predictable process for future AI model launches. OpenAI expects GPT-5.6 to become available to more ChatGPT users, Codex customers, and API developers in the coming weeks.
According to OpenAI, GPT-5.6 Sol is its most capable AI model so far, offering stronger performance in coding, biology, and cybersecurity. It introduces a new “Max” reasoning mode along with an “Ultra” mode that coordinates multiple AI subagents to tackle highly complex tasks.
OpenAI claims Sol slightly outperforms Anthropic’s Claude Mythos 5 on coding workflows while generating only about one-third as many output tokens. The company also says it has significantly strengthened the model’s security by making it more resistant to jailbreak attempts and optimizing it to prioritize defensive cybersecurity guidance over offensive hacking techniques.
Unlike some previous AI safety systems that relied on external content filters, OpenAI says GPT-5.6 has its safety protections built directly into the model itself. This approach is intended to reduce unnecessary false positives while still preventing misuse.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
GPT-5.6 will be offered in three pricing tiers once it becomes generally available. Sol will cost $5 per million input tokens and $30 per million output tokens, Terra will be priced at roughly half that amount, while Luna will cost $1 per million input tokens and $6 per million output tokens. OpenAI also says it has improved prompt caching to make repeated requests more affordable and consistent.
