Microsoft Edge brings new on‑device AI models, translation & speech APIs to the browser

Microsoft is baking a new small language model straight into Edge, and it’s not stopping there. The browser is picking up on‑device translation and speech features too, so a lot more AI work can happen on your PC without running everything through the cloud.

The new model is called Aion-1.0-Instruct. Microsoft describes it as a smaller and faster successor to the Phi-4-mini model it started testing in Edge last year through the Prompt and Writing Assistance APIs. Basically, instead of every web app shipping its own model or calling a remote service, Edge exposes a built-in one that sites can tap into for things like rewriting text, summarizing content or giving writing help.

Right now Aion-1.0-Instruct is only in preview. You need the Canary or Dev channels of Microsoft Edge and a feature flag to play with it. Microsoft says the model is tuned to run on a wider range of hardware, including machines that wouldn’t comfortably handle bigger LLMs that live in the cloud or in separate apps. That fits neatly with what the company has been selling around Copilot+ PCs and “AI PCs” in general, where local models are supposed to feel instant and not choke your laptop every time you ask for a rewrite.

Edge is also getting two new task-specific models that live in the browser. There is a Language Detector API that can figure out what language a given bit of text is using, and a Translator API that can convert between more than 145 languages on the device. Web apps and extensions can call these from JavaScript. That means a site could quietly see that you are typing in Hindi and then, for example, keep a translated English version of your message updated in another field without sending every keystroke to a cloud translation service.

Microsoft also made sure to highlight the privacy and cost angle. If translation happens locally, then there is nothing for a server to log and no per-request billing tied to every little feature in a web app. For developers, that also means one less external dependency to manage. They can rely on the browser’s AI stack as long as the user is on a new enough version of Edge.

On speech, Edge is wiring on‑device speech recognition into the standard Web Speech API, again starting in Canary and Dev builds. If it works the way Microsoft describes it, a website could offer dictation or voice commands that feel closer to a native app, while the browser quietly routes the audio into a local model first instead of immediately pushing it to a server. Microsoft still leaves room for cloud speech services, but the default pitch is faster and more private since the audio does not have to leave the machine for basic scenarios.

We already saw the first part of this story last year when Edge’s Prompt API showed up with Phi-4-mini, turning the browser into a kind of host for small language models that any site could borrow. Now Microsoft is rounding it out with a newer model, translation tools that ship in Edge 148, and speech hooks that are still in preview.

That said, as Microsoft and Google bake more AI into their browsers, alternatives like Vivaldi are winning over users fleeing from the AI rush. Interesting times ahead!

We stand out from the tech-media crowd because we break news stories; we mainly bring you stuff that you won’t find anywhere in the mainstream tech media. Our stories have been picked up by some of the world’s most popular websites and media outlets—more info is available here.