Official documentation
Phi Silica in the Windows App SDK
Microsoft Learn, 2026
Read the originalRead on October 16, 2026
- What it is
- Microsoft's developer documentation for Phi Silica, the small language model that Windows ships for on-device language features, reached through the Windows App SDK. The page says the model is optimized for Copilot+ PCs, where it runs on the neural processing unit, and that it is being replaced by Aion Instruct.
- How much weight it carries
- Official Microsoft documentation, so it is the authority on the model's hardware and status. It is written for developers, describes an experimental API, and changes often, so check the live page before relying on dates or version details. It does not describe fluid dictation directly.
Facts taken from this source
- Microsoft says the Phi Silica APIs are part of a Limited Access Feature.
“The Phi Silica APIs are part of a Limited Access Feature” (Important note under the page heading)
- Microsoft's developer page says Phi Silica is being replaced by Aion Instruct, a new on-device model.
“Phi Silica is being replaced by Aion Instruct” (Important note at the top of the page)
- Microsoft says its Phi Silica language model is optimized for Copilot+ PCs, where it runs on the neural processing unit.
“Phi Silica is optimized for efficiency and performance on Windows Copilot+ PCs (where it runs on the NPU)” (Introduction)
- On NPU-equipped devices, Phi Silica uses speculative decoding, with a smaller draft model, to speed up text generation.
“employs a technique called speculative decoding to accelerate text generation using a smaller draft model” (Introduction, paragraph on how the model runs)