Microsoft launches MAI-Code-1.1-Flash AI for programmers

Microsoft has introduced MAI-Code-1.1-Flash, a new artificial intelligence model specifically designed for programming tasks within GitHub Copilot. This compact and efficient model serves as a replacement for the version presented during the company’s Build conference earlier this year.

Designed to compete with tools like Claude Code, the new model focuses on command-line workflows and .NET project performance. Microsoft reports that the model is significantly more cost-effective, costing approximately one-fourth of its predecessor to operate, while also being 25% faster at generating tokens. Internal benchmarks indicate a 22% improvement in Terminal-Bench 2.1 tests.

To address legal concerns regarding data usage, Microsoft stated that the model was trained from scratch using its own verifiable information, avoiding the practice of model distillation from third-party sources. Real-world usage data shows a 4% increase in code retention after commits and a 9% rise in developer return visits to the tool.