OpenAI expands Daybreak cybersecurity project with new GPT-5.5-Cyber model

OpenAI is expanding its Daybreak cybersecurity project, offering a limited release of the GPT-5.5-Cyber model to defense teams. The project helps teams find, verify and fix vulnerabilities faster.

OpenAI says the main challenge now is fixing bugs, not finding them. GPT-5.5-Cyber scored 85.6% on CyberGym, beating GPT-5.5's 81.8% and Claude Mythos 5's 83.8%. It also scored 39.5% on ExploitGym, much higher than GPT-5.5's 25.95%, and 69.8% on SEC-bench Pro, versus GPT-5.5's 63.1%.