An article summarized by Quartz:
OpenAI announced three new artificial intelligence models, GPT-5.6 Sol, Terra, and Luna, but is limiting initial access to just 20 government-approved "trusted partners" at the request of the Trump administration. The company said it briefed government officials before the launch and is taking a phased rollout approach while working toward a broader public release in the coming weeks. Amazon's Bedrock platform is one of the ways approved partners can access the models.
The three models are designed for different use cases, with Sol being the most advanced, Terra offering mid-level capabilities, and Luna targeting lower-cost applications. OpenAI said Sol performs especially well in cybersecurity, coding, and biology, particularly in identifying and fixing software vulnerabilities rather than carrying out cyberattacks. While all three models were classified as high-capability under OpenAI's internal safety framework for cybersecurity and biological risks, the company said none crossed its highest "critical" risk threshold.
The limited release comes as the Trump administration develops new oversight procedures for advanced AI systems. Earlier this month, Anthropic was forced to withdraw two of its models after the administration attempted to restrict access based on users' nationality, a policy that proved technically impossible to enforce. OpenAI said it hopes to work with the administration to establish a consistent process for reviewing and releasing future AI models, while emphasizing that long-term government-controlled access should not become the standard because it could slow innovation and limit access for developers, businesses, and security researchers.
