GPT 5.6 System Card (deploymentsafety.openai.com)

OpenAI has announced the launch of GPT-5.6, which features a trio of models: Sol, the flagship with advanced capabilities; Terra, a more accessible option; and Luna, optimized for speed and cost-efficiency. With enhanced safeguards tailored to each model's capabilities, OpenAI is committed to addressing safety and risk, particularly in cybersecurity and biological domains. While the new models demonstrate significant advancements in cybersecurity, they do not reach the highest risk levels. Notably, GPT-5.6 is better at identifying vulnerabilities than executing complex cyber attacks, emphasizing its role as a defensive tool. The rollout of GPT-5.6 includes strict safety measures that block a wider range of harmful activities, with new technologies implemented to intervene in real-time during content generation. OpenAI has engaged extensive testing and continuously assesses the models against various risks, with over 700,000 GPU hours dedicated to identifying vulnerabilities. By granting broader access to GPT-5.6's cybersecurity capabilities, OpenAI aims to empower defenders to patch weaknesses before they can be exploited, fostering a more secure environment in an increasingly digital landscape.