MAI-Cyber-1-Flash in MDASH
Security model from the MAI-Thinking-1 lineage inside the MDASH multi-agent system scores 96% any-crash on CyberGym at half the earlier cost.
A compact vulnerability-finding model plus agents that find, validate and fix flaws; Microsoft says it beats the prior MDASH setup built on GPT-5.4, mini and 5.3 Codex at 50% lower cost.
- Date
- Thursday, 13 August 2026
- Lab
- Microsoft
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| CyberGym (MDASH + MAI-Cyber-1-Flash) | 96% any-crash; 86.3% final leaderboard company evaluation | company |
Sources
This record was checked against its sources on 6 October 2026. How we check