TL;DR: Microsoft released two small-parameter text LLMs: MAI-Thinking-1 (35B reasoning) and MAI-Code-1-Flash (5B code), with the latter rolling out to GitHub Copilot users in VS Code.
Summary: Microsoft announced MAI-Thinking-1, a 35-billion-parameter reasoning model available to select early partners, and MAI-Code-1-Flash, a 5-billion-parameter model built for GitHub Copilot and VS Code to deliver high performance at lower cost. The code model is rolling out to individual GitHub Copilot users in Visual Studio Code.
Why it matters: These low-parameter models signal a shift toward efficient, cost-effective AI inference for developers. AI builders should evaluate these models for on-device or latency-sensitive coding and reasoning tasks.
Source: simonwillison.net