Bullish

Microsoft Cyber Model MAI-Cyber-1-Flash Cuts Costs 50%, Handles 90% Tasks

2026-07-28 09:21:22

Microsoft deploys MAI-Cyber-1-Flash for 90% of security tasks, reducing costs by half versus GPT-5.4. Public testing for Project Perception starts Aug 3.

Woofun AI reports that Microsoft has launched MAI-Cyber-1-Flash, a 137-billion-parameter cybersecurity model integrated into the MDASH vulnerability scanning system. The model handles up to 90% of discovery and remediation tasks, routing the remaining 10% to GPT-5.4, resulting in nearly 50% lower costs compared to existing GPT solutions.

In CyberGym testing, the MAI-Cyber-1-Flash and GPT-5.4 combination achieved a 95.95% score, outperforming GPT-5.5 Cyber and Anthropic Mythos 5. The model will also integrate with Project Perception, which begins public testing on August 3rd, while MAI-Cyber-1-Flash remains in private testing for audited MDASH clients.

WOOFUN AI

Impact Assessment · Quick Read

This development signals a shift toward specialized, cost-efficient AI models for enterprise security operations. By handling the majority of routine tasks with a smaller model, Microsoft demonstrates a scalable approach to reducing inference costs while maintaining high accuracy through hybrid workflows. The upcoming public test of Project Perception may accelerate adoption of AI-driven security agents across the industry.
Generated by WOOFUN AI · For reference only, not investment advice

Comments

Me
Replying to @User
0/800

No comments yet.

Notifications

Sign in to view messages
View all messagesManage subscriptions