Microsoft says its latest cybersecurity AI model beats two well-known rivals — Claude Mythos and GPT-5.6 Sol — in a benchmark test. The new system can deploy more than 100 AI agents to hunt for software flaws, and it does so at half the cost of Microsoft's current best configuration.
How the model works
The model is designed to automate vulnerability discovery. Instead of relying on a single AI, it orchestrates a swarm of over 100 specialized agents. Each agent focuses on a different part of the codebase or a specific type of weakness. Together, they cover more ground than a single model could.
Microsoft claims this approach leads to better results than the two competing systems. Claude Mythos and GPT-5.6 Sol are both general-purpose large language models that have been adapted for security tasks. But Microsoft's dedicated cybersecurity model reportedly finds more flaws with less computational overhead.
Cost and performance
The company says the new model cuts costs by 50% compared to its previous top-tier setup. That's a significant reduction for organizations that run continuous security testing. Lower costs could make advanced vulnerability scanning accessible to smaller firms that previously couldn't afford it.
Microsoft hasn't released detailed benchmarks or independent validation yet. The claims are based on internal testing. Still, the company is positioning the model as a major step forward in AI-driven security.
If the performance holds up under outside scrutiny, the model could change how companies approach bug hunting. Automated agents that work in parallel could find vulnerabilities faster than human teams or single AI systems. That speed matters when attackers are constantly probing for weak points.
But the announcement also raises questions. How do you verify that an AI agent hasn't missed a critical flaw? And what happens if the agents themselves introduce new vulnerabilities? Microsoft hasn't addressed those concerns yet.
The company has not said when the model will be available to customers or whether it will be integrated into existing products like Azure Security Center. For now, the claim stands as a bold statement in a competitive field.
