**[OpenAI Launches GPT-5.5-Cyber, Surpasses Anthropic in Benchmark Testing]** According to Deep Tide TechFlow, on June 24, Decrypt reported that OpenAI officially launched the GPT-5.5-Cyber cybersecurity model on June 22 as part of the Daybreak Network Defense Program. The model achieved a score of 85.6% in the CyberGym benchmark test developed by UC Berkeley, surpassing Anthropic's Mythos 5 (83.8%) and Claude Opus 4.7 (73.1%). Anthropic's Mythos 5 and Fable 5 were taken offline due to an emergency export control directive issued by the U.S. government on June 12. GPT-5.5-Cyber is only accessible to verified security professionals. OpenAI has established partnerships with seven countries, including Australia and Canada, as well as 28 security companies. Its Codex Security tool, launched in March, has scanned over 30 million code submissions and fixed more than 500,000 vulnerabilities.