OpenAI announced on Friday that it is restricting the release of its new GPT-5.6 model family to a small group of trusted partners, following a request from the Trump administration. The move affects the flagship model Sol, the balanced Terra, and the faster low-cost Luna, and highlights growing tensions between AI developers and government oversight.
Three models under restriction: Sol, Terra, and Luna
The GPT-5.6 lineup features Sol with advanced agentic capabilities in coding, biology, and cybersecurity. Terra offers a balanced everyday alternative, while Luna delivers speed at lower cost. Despite their promise, all three are limited to partners whose participation has been shared with the government. OpenAI emphasized this is a temporary step to foster dialogue on responsible AI deployment.
The shadow of Anthropic's Fable 5 and regulatory concerns
The request follows the earlier case of Anthropic's Fable 5, which was taken down after the government ordered removal of access for foreign nationals. Former White House AI adviser Dean Ball, soon to join OpenAI, criticized the current executive order as a de facto involuntary licensing regime. Without clear safety standards, the government may cause indefinite launch delays, potentially ceding ground to China in the AI race and jeopardizing billions in infrastructure investments.
Sponsored Protocol
Sol's performance and built-in safety measures
OpenAI claims Sol outperforms Anthropic's Claude Mythos 5 in coding benchmarks while using one-third of the output tokens. Sol integrates the strongest security stack ever, with guardrails embedded directly into the model's core behavior rather than external filters. This avoids the false positives that plagued Fable 5, which routed sensitive queries to older models. Sol is optimized for defensive cybersecurity, prioritizing protection over exploitation.
Sponsored Protocol
Pricing and planned broader availability
Pricing is tiered: Sol at $5 per million input tokens and $30 per million output; Terra at half; Luna at $1 and $6 respectively. Improved prompt caching reduces costs for repeated queries. OpenAI expects to extend access to ChatGPT, Codex, and the API in coming weeks, while working with the administration on cybersecurity rules and a repeatable release process. The situation mirrors recent API restrictions in Europe, where Italian SMEs faced similar challenges, underscoring the need for transparent global standards.
