GitHub

IT House News, June 3, at the Build 2026 Development Congress, which opened today, Microsoft announced last year the launch of a writing aid based on the Phi-4-mini model for the Edge Browser API Based on this, it expands its end-side AI capability by adding models and API. This update consists of three main elements:

Aion-1.0-Instract Small Language Model developer preview version (for early testing and feedback);

(a) Language testing and translation of API in Edge 148 versions, driven by end-to-end task-specific models;

And the experimental Web Speech API-side speech recognition function provided in the Edge Canary and Dev channels.

Microsoft indicates that in the past year, the Edge Browser's writing aids API have been based on the Phi-4-mini model. This is a model of 4 billion parameters, which is excellent in terms of text interpretation, reasoning and compliance with directives, but its hardware requirements limit its availability on different equipment.

Therefore, Microsoft immediately introduced a preview of the Aion-1.0-Instract small language model in the Edge Canary and Dev Channels. The model is smaller, faster, more efficient and extends to more equipment - including GPU Lower performance equipment, and through CPU The reasoning supports GPU-free devices while providing good output quality for a wide web-based use.

The preview version allows developers to assess Aion-1.0-Instract in the real Web scene, test and provide feedback on API interoperability, and the model is scheduled to be released as an open source to Hugging Face in July.

New language testing and translation API is now officially available in the Edge 148 preview. These APIs allow sites and browsers to expand the recognition of text language and to translate between languages, support more than 145 languages based on end-to-end task-specific models and optimize the Web translation load.

Developer can use on website or extension JavaScript Call these APIs, which provide better user privacy, network independence and zero translation costs than cloud services.

In the latest Edge Canary and Dev Channels, Microsoft also introduced a side mission-specific model for voice processing, enabling local voice recognition for Web Speech API. This will result in the local completion of the voice-to-text process on user devices, which will improve user privacy, reduce delays and support low-network connectivity or no-network scenes. Developer needs only a few changes in the existing Web Speech API code, e. g. settings

= True, you can enable end-side voice recognition. Microsoft indicates that with the Aion-1.0-Instract small language model, new language testing and translation API and end-side voice recognition, developers can use models embedded in browsers to build an AI-based Web experience without relying on specialized hardware, cloud services or field-specific expertise.

Advertising statements: The external jump links (including not limited to hyperlinks, 2D codes, passwords, etc.) contained in the text are used to convey more information and save time for selection purposes only for reference purposes, which are included in all IT House articles.