Abishek D Praphullalumar
Microsoft is expanding DeepSeek R1 AI models to Windows, integrating distilled versions into Copilot+ PCs. Initially available on Snapdragon X-powered devices, the models will later support Intel Core Ultra 200V and AMD Ryzen AI 9 chips.
The first model, DeepSeek-R1-Distill-Qwen-1.5B, will soon be joined by larger 7B and 14B versions, downloadable via Microsoft’s AI Toolkit. Optimized for NPUs, these models balance CPU and NPU workloads for efficiency, achieving 130ms first-token latency and 16 tokens per second for short prompts.
Despite its deep investment in OpenAI, Microsoft continues expanding AI diversity, hosting DeepSeek alongside GPT, Llama, and Mistral in its Azure AI Foundry. Developers can access these models via VS Code’s AI Toolkit for local testing. Model distillation enables powerful AI performance on consumer hardware, making advanced AI more accessible.
