Microsoft Launches MAI-Cyber-1-Flash for Cybersecurity Workflows

On July 27, 2026, Microsoft introduced MAI-Cyber-1-Flash, a model designed for cybersecurity workflows within the MDASH multi-agent platform. This model aims to handle 90% of vulnerability analysis tasks, while the remaining 10% are managed by OpenAI's GPT-5.4. The combined system reportedly achieved a 95.95% success rate on CyberGym, although this result was not yet reflected on CyberGym's public leaderboard. Microsoft claims that the new model cuts costs by 50% compared to previous configurations. However, the launch materials did not clarify the routing distribution of tasks between the models. The model card also includes various performance metrics on different benchmarks, but these figures are not directly comparable to the CyberGym result. The launch of MAI-Cyber-1-Flash signifies a significant advancement in Microsoft's cybersecurity capabilities.

Key Points: • MAI-Cyber-1-Flash can handle up to 90% of cybersecurity tasks within MDASH. • The combined system achieved a 95.95% success rate on CyberGym, though not yet reflected publicly. • The new model reduces operational costs by 50% compared to previous configurations.