Microsoft Integrates DeepSeek R1 into Copilot+ PC: What Does This Mean for Users?

The artificial intelligence DeepSeek has quickly established itself in the mobile sector and is now making its way into the Windows world. Microsoft is actively supporting this process by integrating the DeepSeek R1 model into its cloud platform, Azure AI Foundry. Now, the company is taking things a step further by announcing that distilled versions of this model will be available on Copilot+ PCs.

What Are Distilled DeepSeek R1 Models?

A distilled model is a simplified version of a large neural network that retains essential capabilities while operating more quickly and efficiently. The full version of DeepSeek R1 consists of 671 billion parameters, while the distilled models will range from 1.5 billion to 14 billion parameters. This allows them to run directly on consumer devices without requiring expensive specialized hardware.

The first available model will be DeepSeek-R1-Distill-Qwen-1.5B (with 1.5 billion parameters), with more powerful versions of 7B and 14B to follow soon. These models will be accessible via the Microsoft AI Toolkit and will initially run on devices with Snapdragon X chips and Intel Core Ultra 200V processors, later expanding to AMD Ryzen AI 9.

How Is Microsoft Optimizing These Models for Copilot+ PC?

To ensure maximum performance on PCs, Microsoft has adapted these models for neural processing units (NPUs). The most critical computations, such as the transformer block, are executed on the NPU, while memory-intensive tasks are handled by the central processor.

Thanks to this optimization, Microsoft has achieved a response time of 130 milliseconds for the first token and a processing speed of 16 tokens per second (for short inputs of up to 64 tokens). This means that users will receive AI-generated responses almost in real-time—even without a cloud connection.

Microsoft Expands Its AI Ecosystem

Although Microsoft works closely with OpenAI, its strategy clearly leans toward diversification. Its Azure Playground platform already features models from GPT (OpenAI), Llama (Meta), and Mistral—and now DeepSeek R1. This indicates that Microsoft aims to create an open AI ecosystem where users can choose which model to use.

DeepSeek R1 on Copilot+ PCs is another step toward integrating powerful AI tools into everyday devices. The only question that remains is: how effective will this technology be in real-world use? We’ll find out soon enough.