Microsoft’s Aion Instruct is now generally available: a 2.3-billion-parameter open-weights model that replaces Phi-4-mini as the built-in AI inside Microsoft Edge — and, through a new Prompt API origin trial, any website can now run it on a visitor’s own device. Why it matters: AI features are starting to ship inside the browser itself, with no server bill and no data leaving the machine.
Redmond, Washington — November 19, 2026
What shipped
The Microsoft Edge team announced three updates on November 19. First, Aion Instruct reached general availability, both as an open-weights download on Hugging Face (listed as microsoft/Aion-1.0-Instruct-2.3B, MIT-licensed per its model card) and as Edge’s built-in small language model. Second, the Prompt API opened an origin trial, so developers can call it from live websites rather than only in the Canary and Dev test channels. Third, an experimental Decisions API entered developer preview in Edge Canary and Dev.
The model itself
At 2.3B parameters, Aion Instruct is smaller than the Phi-4-mini it replaces. Microsoft says it is faster, runs on many more devices, and delivers comparable quality on common web tasks: text synthesis and generation, summarization, instruction following and structured outputs. The GA release follows a developer preview of Aion-1.0-Instruct; Microsoft says it refined the training using early-adopter feedback.
Edge downloads the model only when a site or feature first needs it, and only on devices that can run it well. Users stay in control: on-device AI can be turned off in Edge settings, which removes the model and stops future downloads. Microsoft says Aion is also coming to Windows later this year, laying the foundation for on-device AI across Windows and Copilot+ PCs.
Two APIs for developers
The Prompt API is the bigger idea. Microsoft calls the approach “hybrid intelligence”: a site uses the on-device model when it is available and well-suited, and switches to a configured cloud provider when it needs broader device coverage or more powerful inference. Microsoft is asking for feedback on interoperability as the API is considered for standardization with other browser vendors.
The Decisions API, powered by Aion, answers yes/no, multiple-choice or scoring questions about a piece of text in a single pass — returning probabilities and a confidence value without generating any text. Microsoft’s example: reading a product review once and reporting what it is about, how satisfied the customer is, and whether they are asking for help — the kind of fast classification work sites currently send to a cloud model.
Why it matters
A free, capable model sitting inside one of the world’s major browsers changes the math for web developers: summaries, writing help and content routing can run locally, instantly, and without sending user text to a server. It is also a quiet competitive move — whoever owns the browser’s built-in model owns the default AI layer of the web. For users, the honest summary is: your browser just got an AI upgrade, it runs on your hardware, and you can switch it off.
Frequently asked questions
What is Aion Instruct?
Microsoft’s 2.3B-parameter small language model for on-device AI. It is now generally available as Edge’s built-in model and as an open-weights download on Hugging Face.
Does it send my data to the cloud?
Aion runs on your device. Websites can optionally fall back to a cloud provider they configure when the local model is unavailable or unsuitable — that choice is the site’s, and Edge lets you disable on-device AI entirely in settings.
When is it coming to Windows itself?
Microsoft says later this year, as groundwork for richer on-device AI experiences across Windows and Copilot+ PCs. No specific date was given.
Sources: Microsoft Edge Blog, Hugging Face, Windows Forum
