Optimizing the frontier performance curve | Microsoft AI

Microsoft released MAI-Cyber-1-Flash, which topped the CyberGym benchmark at 50% cost while running on H100s.

• MAI-Cyber-1-Flash beat Mythos by 12 percentage points on CyberGym.
• It handles up to 90% of tasks, reserving GPT 5.4 for the hardest 10%.
• MAI-Code-1-Flash shows 10% higher code accept rate and 10% lower token usage than GPT-5.4 Mini and Claude Haiku 4.5 in VS Code.
• MAI-Image-2.5-Flash cuts PowerPoint GPU costs up to 84% versus GPT-Image-2.
• MAI-Voice-2-Flash reduces Dynamics 365 Contact Center GPU costs by up to 89%.

Enterprises running Microsoft AI products can now get frontier-level performance at a fraction of the GPU cost, changing their cost-performance tradeoffs.

Sources

• Microsoft AI: Optimizing the frontier performance curve | Microsoft AI

Read this as text

Back to the AI news