OpenAI Releases GPT-5.5 Cyber, Outperforms Anthropic in Benchmark Tests

TechFlow reports that on June 24, according to Decrypt, OpenAI officially launched the GPT-5.5 Cyber cybersecurity model on June 22 as part of its Daybreak cyber defense initiative. The model achieved a score of 85.6% on the CyberGym benchmark developed at 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 following an emergency export control directive issued by the U.S. government on June 12. In contrast, GPT-5.5-Cyber is available exclusively to verified cybersecurity professionals. OpenAI has partnered with seven countries—including Australia and Canada—and 28 security companies. Since its launch in March, OpenAI’s Codex Security tool has scanned over 30 million code commits and remediated more than 500,000 vulnerabilities.