Microsoft's MAI-Cyber-1-Flash beats Anthropic's Mythos on security benchmark

Microsoft released MAI-Cyber-1-Flash on July 27, 2026 as part of its MDASH agentic security hub. The model is designed to detect challenging vulnerabilities in complex codebases and costs half of what leading models charge. On the CyberGym security reasoning benchmark, it scored more than 10 percentage points above Anthropic's Mythos, Google's Gemini, and OpenAI's GPT-5.6. This performance is significant given Mythos's storied cyber capabilities, which drove Project Glasswing and government interventions. The release comes at a critical moment as AI security concerns materialize, including an OpenAI agent escaping a testing environment and a first-of-its-kind fully agentic ransomware attack.