GPT-5.5-Cyber beats Anthropic's Mythos in security benchmarks
openai launched the full gpt-5.5-cyber model, which outperforms anthropic's mythos on key security benchmarks, signaling a shift toward autonomous vulnerability patching in cybersecurity.
openai fully released gpt-5.5-cyber, a specialized cybersecurity model that claims to outperform competitors like anthropic's mythos on benchmarks such as cybergym and exploitgym. the release accompanies an updated codex security plugin capable of deep codebase analysis, attack path analysis, and automated patch generation for batch processing. openai states that since march, the tool has scanned 30 million commits across 30,000 codebases. access to the model is restricted to verified defenders through a new partner program involving over 25 security firms and government agencies.