OpenAI has limited the release of its GPT-5.6 models to a 'small group of trusted partners' at the request of the U.S. government, the company announced Friday. The next-generation lineup includes Sol, its flagship model; Terra, a more balanced model for everyday use; and Luna, a faster, lower-cost option. The Trump administration has restricted the release of all three models, prompting OpenAI to limit preview access to partners whose participation has been shared with the government. The decision comes amid increased government pressure on AI companies to restrict their most advanced systems, as seen in the case of Anthropic's Fable 5, which was taken offline after the administration ordered its removal for foreign access.
The administration’s request has raised concerns about the extent of government control over AI model releases. Dean Ball, a former White House AI advisor and soon-to-be OpenAI employee, argues that President Trump’s executive order, which requires certain AI companies to voluntarily submit their most advanced models for government review up to 30 days before release, has created a de facto involuntary licensing regime for frontier AI. Ball warns that without clearly defined safety standards, the process could lead to endless launch delays, potentially benefiting China in the AI race and jeopardizing billions in AI infrastructure investments. OpenAI, while complying with the request, expressed dissatisfaction with the arrangement, stating that the government access process should not become the long-term default.
OpenAI said the preview is a 'short-term step' toward broader availability, as the company works with the administration to develop a new executive order framework on cybersecurity and a 'repeatable process for future model releases.' GPT-5.6 Sol, the company’s strongest model, features improved agentic capabilities in coding, biology, and cybersecurity. It includes a 'max' reasoning effort mode and an 'ultra' mode that uses coordinated subagents for complex tasks. OpenAI claims Sol performs slightly better than Anthropic’s Claude Mythos 5 in coding workflows and uses a third of the output tokens. The firm also emphasized that Sol includes its most robust security stack, optimized for defensive cybersecurity over offensive exploits. It is designed to be hard to jailbreak while prioritizing user education on defense against exploits. OpenAI also noted that its safety guardrails are integrated into the core model’s behavior, rather than relying on external filters. The firm is likely avoiding the pitfalls that affected Anthropic’s Fable 5, which was routed to older models for high-risk topics, leading to user backlash.
Source: techcrunch
