Microsoft overhauls MDASH with MAI-Cyber-1-Flash, delivering broad cybersecurity AI at half the cost
Microsoft announced MAI-Cyber-1-Flash, a cybersecurity-focused model inside MDASH that can handle about 90% of MDASH tasks, with GPT-5.4 tackling the hardest 10%, and claims CyberGym performance of 95.95% at roughly half the cost of the current MDASH model mix. Access is limited to approved MDASH customers via an Azure AI Foundry private preview. MAI-Cyber-1-Flash is a 137-billion-parameter sparse Mixture-of-Experts model with a 256k-token context, designed to replace about 80% of existing MDASH models, and is evaluated in isolated settings with various benchmarks; results may not be directly comparable to public leaderboards. Project Perception will expand the security automation workflows with a public preview planned for August 3.
