Microsoft $MSFT Launches MAI-Code-1.1-Flash, a Compact Coding Model...
Microsoft $MSFT Launches MAI-Code-1.1-Flash, a Compact Coding Model Built to Run On-Device
Microsoft released MAI-Code-1.1-Flash, a small code-generation model designed to run locally on-device, targeting hardware with over 120GB of RAM — like the new RTX Spark chip with its 128GB unified memory.
This builds on MAI-Code-1-Flash, first introduced at Build 2026 back in June. The 1.1 update delivers 25% better token efficiency and cuts running costs to a quarter of the original version. On Microsoft's Terminal-Bench 2.1 results, it shows a 22% improvement in GitHub Copilot CLI performance and 15% better results on .NET-related tasks, responding to developer requests for better real-world task quality.
The model's size dropped 80% from the previous version via 3-bit precision compression, while supporting a 256K context window. MAI-Code-1.1-Flash is now live on GitHub Copilot — through the app, CLI, and directly in VS Code.