OpenAI Releases GPT-5.6 Family with Government-Controlled Access Restrictions
Tags AI · Policy & Law · Enterprise
OpenAI previewed a three-tier GPT-5.6 model family—Sol (top tier), Terra (mid tier), and Luna (fast/economic tier)—with initial access limited to approximately 20 U.S. government-approved organizations. GPT-5.6 Sol achieves 91.9% on Terminal-Bench 2.1 and features 'ultra mode' that spawns subagents for multi-step tasks. Security safeguards block biology, chemistry, and cybersecurity information, requiring multi-agent coordination for complex requests. Pricing ranges from $1-5 per 1M input tokens depending on tier.
Technical significance
First major AI model release with prior government pre-approval, establishing a precedent for national security screening of frontier AI. The three-tier pricing structure and security safeguards indicate a mature enterprise-focused product strategy. Blocks on dangerous domains suggest regulatory compliance is becoming a product feature rather than afterthought.