Home / AI news / Microsoft ships Aion Instruct as Edge's built-in on-device model and opens a Prompt API origin trial
AI news

Microsoft ships Aion Instruct as Edge's built-in on-device model and opens a Prompt API origin trial

Microsoft said on November 19, 2026 that Aion Instruct, a 2.3B-parameter small language model, is generally available as open weights on Hugging Face and as Microsoft Edge's built-in model, and that websites can now test Edge's on-device Prompt API through an origin trial.

What happened

  • Aion Instruct replaces Phi-4-mini as Edge's built-in small language model; Microsoft says it is smaller and faster and runs on more devices while giving comparable quality for summarizing, generating text, following instructions and structured output.
  • The model is available as open weights on Hugging Face and, per Microsoft, is coming to Windows later this year.
  • A Prompt API origin trial lets developers call the on-device model from their live sites in Edge; until now the API was only in Edge Canary and Dev as a developer preview.
  • An experimental Decisions API, in developer preview in Edge Canary and Dev, returns probabilities and a confidence value for yes/no, multiple-choice or scoring questions about text without generating text.
  • Edge downloads the model only when a site or feature first needs it and only on capable devices; users can turn off on-device AI in settings, which removes the model.
Why it matters: Web developers can add private, no-API-cost AI features like summarizing or routing user text that run on the visitor's device in Edge, falling back to a cloud model when needed. The APIs are still trials or previews, so expect changes before standardization.

Sources

Checked November 20, 2026 against the sources above. Claims about performance or benchmarks are the company's own unless stated otherwise.

More AI news