Microsoft Says New Cybersecurity AI Model Helps MDASH Score 95.95% at Half the Cost

Key Points:

• Microsoft has introduced MAI-Cyber-1-Flash, its first cybersecurity-specific model, integrated within MDASH, a multi-model vulnerability identification and remediation system, achieving a 95.95% score on the CyberGym benchmark.
• MAI-Cyber-1-Flash handles up to 90% of MDASH tasks, with the more powerful GPT-5.4 reserved for the most complex 10%, resulting in a reported 50% cost reduction compared to Microsoft's prior best MDASH model combination.
• The 95.95% CyberGym score reflects the combined MDASH system performance rather than the new model alone, and the result has not been publicly listed or independently verified on CyberGym's leaderboard.
• MAI-Cyber-1-Flash is a large sparse mixture-of-experts transformer with 137 billion parameters, designed as a cybersecurity fine-tune of previous models, and replaces 80% of MDASH's existing models to improve efficiency and performance.
• This launch marks the first announced use case of Microsoft's broader Project Perception security system, which will enter public preview in August 2026 and aims to expand beyond vulnerability management to other security workflows.