Microsoft Ships Aion Instruct, a Small On-Device Model for the Web, and Opens the Prompt API to Live Sites
Microsoft Ships Aion Instruct, a Small On-Device Model for the Web, and Opens the Prompt API to Live Sites
Microsoft has made Aion Instruct generally available as an open-weights model and as the built-in small language model in Microsoft Edge, the company announced on November 19, 2026. The release also opens an origin trial for the Prompt API, letting developers call on-device AI from live websites for the first time, and adds a developer preview of an experimental Decisions API.
Aion Instruct replaces Phi-4-mini as Edge’s built-in small language model. At 2.3 billion parameters, Microsoft describes it as smaller, faster, and able to run on many more devices than its predecessor, while delivering comparable quality for common web tasks such as text synthesis and generation, summarization, instruction following, and structured outputs. The model is available as open weights on Hugging Face, and Microsoft says it is coming to Windows later this year as the foundation for richer on-device AI experiences across Windows and Copilot+ PCs.
From preview to general availability
Aion Instruct follows an earlier developer preview of Aion-1.0-Instruct. According to Microsoft’s announcement, the Edge team incorporated feedback from early adopters and refined the model’s training to improve quality and performance for on-device AI on the web before taking it to general availability.
Edge downloads Aion Instruct only when a site or feature first needs it, and only on devices that can run it well. Users stay in control: they can turn off on-device AI in Edge settings, which removes the model and stops future downloads.
The Prompt API moves to live sites
The second update is the Prompt API origin trial. Until now, the Prompt API was available only as a developer preview in Edge Canary and Dev channels. With the origin trial, developers can call the API from a live site in Microsoft Edge and test on-device AI features with a much broader audience.
The origin trial lets developers bring on-device AI features to their sites, evaluate them under real-world conditions, and refine their implementations based on actual usage. Microsoft is also asking developers to share feedback on the API design, the model’s performance, and interoperability challenges across different browsers and models, as the Prompt API is considered for standardization by the web community.
To get started, developers register for the Prompt API Origin Trial, read the developer documentation, try a playground sample, and join the discussion on GitHub.
An experimental Decisions API
The third update is a developer preview of the experimental Decisions API, powered by Aion Instruct and available in Edge Canary and Dev channels for prototyping. Microsoft cautions that both the API and its implementation may change as the company learns from developer feedback.
The Decisions API lets sites ask yes or no, multiple-choice, or scoring questions about a piece of text and receive probabilities and a confidence value, computed in a single pass without any text generation. Microsoft gives examples such as understanding a user’s intent, sorting and routing user-generated content, or scoring and ranking items in real time, all with a few lines of client-side JavaScript.
Why it matters
Microsoft frames the three updates as a step toward what it calls hybrid intelligence on the web: developers can use on-device AI models when they are available in the browser and well suited to the task, and switch to a configured cloud provider to extend device coverage or access more powerful inference when needed.
For website developers, the practical change is that on-device AI is no longer limited to experimental browser builds. A small model that runs locally can handle writing assistance, summarization, language detection, translation, and speech recognition without sending user text to a server, which can mean lower latency, lower inference cost, and stronger privacy for suitable tasks. The Prompt API origin trial is the mechanism that makes that usable on production sites, and the feedback gathered during the trial will shape whether the API becomes a web standard.
The trade-off is capability. A 2.3 billion parameter model is designed for common, well-scoped tasks, not for the hardest reasoning work, and Microsoft is explicit that cloud inference remains the fallback when a task needs more power. The Decisions API is the most experimental piece of the package, available only in Canary and Dev, so developers should treat it as a prototyping tool for now rather than a production dependency.
For users, the change is mostly invisible: Edge will download the model only when a site or feature needs it, and it can be switched off in settings. For developers building on the web, though, November 19 marks the point where on-device AI in Edge moved from preview to something they can ship.
