Microsoft (MSFT) Launches a New Small Language Model Called Mu

Tech giant Microsoft (MSFT) has launched a new small language model called Mu that is built to handle complex language tasks efficiently on devices like Copilot+ PCs. Unlike larger AI models that run on the cloud, Mu works entirely on a device’s Neural Processing Unit. This setup allows it to respond quickly while using less power and memory. The model was developed using insights from Microsoft’s earlier Phi models and was trained on high-quality educational data. It was also fine-tuned using advanced techniques like distillation and low-rank adaptation in order to perform well despite having fewer parameters.

What makes Mu especially efficient is its encoder–decoder architecture, which splits how it processes input and output, unlike typical models that handle both at once. This design, along with smart features like rotary positional embeddings, grouped-query attention, and dual LayerNorm, helps Mu run faster and with less delay. Microsoft also used quantization, which is a method that turns complex calculations into simpler ones by converting model weights from floating-point to smaller integers. This reduces memory use and boosts speed without sacrificing accuracy, therefore making Mu ideal for on-device performance.

To power the AI agent in Windows Settings, Microsoft fine-tuned Mu using more than 3.6 million examples to improve its ability to understand and act on hundreds of system settings. The firm tackled tricky situations—like vague commands involving multiple monitors—by focusing on the most commonly used settings. Short or unclear queries still use regular search, while clearer, multi-word commands activate the AI agent. As a result, Mu delivers fast, accurate help with system settings, all while running on a compact model that’s just one-tenth the size of similar AI tools.

Is MSFT Stock a Buy?

Turning to Wall Street, analysts have a Strong Buy consensus rating on MSFT stock based on 30 Buys and five Holds assigned in the last three months. Furthermore, the average MSFT price target of $516.14 per share implies 6% upside potential.

See more MSFT analyst ratings