.svg){kind=spacer}
Details
- Microsoft has introduced DeepSeek R1 models with 7 billion and 14 billion parameters for Copilot+ PCs, initially available on Qualcomm Snapdragon X devices and slated to expand to Intel Core Ultra 200V and AMD Ryzen systems.
- This release is part of a collaboration between Microsoft’s Azure AI Foundry and hardware partners Qualcomm, Intel, and AMD, aiming to empower developers to create edge AI applications on next-gen laptops.
- The models leverage 4-bit quantization through Microsoft’s Aqua toolchain and Phi Silica optimizations to decrease memory consumption, enabling continuous neural processing unit (NPU) inference at over 40 trillion operations per second (TOPS) with low battery usage.
- The launch builds on Microsoft’s previous 1.5 billion parameter model introduced in January 2025, now offering three times the model capacity; the new 14B model generates tokens at 8 per second, compared to 40 per second for the 1.5B version.
- These advancements support new hybrid cloud-edge workflows, providing local fine-tuning and retrieval-augmented generation (RAG) capabilities, with Microsoft planning further performance optimizations.
Impact
This initiative strengthens Microsoft’s position in the push for advanced on-device AI, offering a viable alternative to Apple’s integrated solutions and Google’s cloud-centric model. By making powerful models available on mainstream hardware, Microsoft accelerates personalized and real-time AI features without heavy cloud reliance. The move underscores a broader industry shift toward hybrid AI and positions Azure as a pivotal bridge between cloud and edge computing.
