Microsoft has launched its first cybersecurity-specific model within MDASH, its multi-model vulnerability identification and remediation system. The new model, named MAI-Cyber-1-Flash, in conjunction with GPT-5.4, has reportedly achieved a score of 95.95% on CyberGym, a known-vulnerability reproduction test.
Performance and Cost Efficiency
According to Microsoft, this configuration is 50% less expensive than its previous best MDASH combination, which included GPT-5.4, GPT-5.4 mini, and GPT-5.3 Codex. Access to this model is currently limited to approved MDASH customers through a private preview via Azure AI Foundry.
Model Capabilities and Design
MAI-Cyber-1-Flash is designed to manage up to 90% of MDASH tasks, with GPT-5.4 reserved for the more complex remaining tasks. It is important to note that this model is not available as a standalone public model or general-purpose API.
The reported score of 95.95% pertains to the MDASH system’s performance using both MAI-Cyber-1-Flash and GPT-5.4, rather than the new model alone. CyberGym Level 1 assesses an agent’s ability to produce a working proof of concept from a vulnerability description and unpatched source code, but it does not evaluate blind vulnerability discovery or the correctness of generated patches.
Benchmarking and Limitations
As of July 28, 2026, Microsoft’s score was not listed on CyberGym’s public leaderboard, which showed Wiz’s Atlas agent leading with a score of 90.9%. Microsoft’s previous submission on May 12 had a score of 88.4%. The comparison of these scores is complicated by the lack of clarity on whether the criteria for the July score are consistent with prior results.
The model card for MAI-Cyber-1-Flash indicates that it is a sparse mixture-of-experts transformer with 137 billion total parameters and five billion active parameters. It has a context window of 256,000 tokens and replaces 80% of MDASH’s existing models, improving the CyberGym result from 88.4% to 95.95%.
Future Developments
This launch is part of Project Perception, which aims to coordinate defensive security agents. Microsoft plans to extend the capabilities of MAI-Cyber-1-Flash beyond software vulnerability management to other security workflows. A public preview of Project Perception is set to begin on August 3.
Microsoft has cautioned that the generated text and code from the model may be inaccurate or incomplete and should be reviewed before any consequential use.
This article was produced by NeonPulse.today using human and AI-assisted editorial processes, based on publicly available information. Content may be edited for clarity and style.
