Microsoft Says New Cybersecurity AI Model Helps MDASH Hit 95.95% at Half the Cost
What happened
Microsoft introduced a new cybersecurity-specific AI model inside MDASH, its multi-model system for vulnerability detection and remediation. The updated MDASH combines the new MAI-Cyber-1-Flash model with GPT-5.4. This setup scored 95.95% on CyberGym, a cybersecurity benchmark, while reducing costs by about 50% compared to Microsoft’s previous best combination that used GPT-5.4, GPT-5.4 mini, and GPT-5.3 Codex. Access to this configuration is limited to approved users.
Why it matters
Lowering costs by half while improving accuracy to nearly 96% is a tangible efficiency gain for cybersecurity operations. Many organizations struggle with the high expense and complexity of fine-tuning AI models for detecting vulnerabilities and automating fixes. MDASH’s new AI blend means security teams can achieve better detection and response fidelity without doubling their compute or licensing budgets. The selective access suggests Microsoft is still controlling rollout, so wide availability and integration into mainstream cybersecurity workflows may take time.
What to watch next
It will be important to see if Microsoft expands access beyond approved users and how this affects wider adoption of AI-driven vulnerability management tools. Watch for updates on how well the new model performs in real-world scenarios outside of CyberGym, especially under diverse and evolving threat conditions. Also, competitors may respond with cost-efficient models to maintain pressure in the cybersecurity AI market.
AI Quick Briefs Editorial Desk
