From the source
Microsoft shared early results for MAI models specialized for agentic workloads in
GitHub Copilot and Excel.
The MAI model deployed in Excel is on par with GPT-5.6 for the most common tasks while being more cost-efficient, and MAI-Code-1-Flash in GitHub Copilot outperforms similarly sized models with a roughly 10% higher code accept rate and 10% lower median token usage than GPT 5.4 Mini and Claude Haiku 4.5.
Microsoft is extending this hill-climbing approach to train efficient models across Copilot Chat, Outlook, PowerPoint, and more.




