Opera proposes WebAI, a standard for AI models running in the browser
A new opt-in API lets sites and extensions call local, on-device models for text generation, classification and embeddings, no API key or server required.
Opera shipped an early preview of WebAI on 27 October, a proposed web standard that lets websites and extensions run AI models locally in the browser instead of calling a cloud API. It’s backed by Google and Microsoft as a standard proposal, and lands first in Opera One, Opera GX and Opera Neon.
Three APIs are on offer: Prompt, for text generation; Decision, for classifying content; and Semantic Embeddings, for relating pieces of text. Models run on Opera’s own inference engine or on compatible models already on the device, and nothing downloads until a user opts in from Settings, per site. For now it only works on 64-bit Windows (Intel or AMD) and Apple silicon Macs: no Linux, no Windows on Arm, no Intel Macs.
// conceptual shape of the proposed API
const result = await ai.prompt.generate('summarise this page');Why it matters: if you’ve been paying per request for simple classification or short text generation, a local model removes the API bill and the network round trip for users on supported hardware, as long as you accept it won’t run everywhere yet.
The caveat: this is Opera’s own implementation of a standard still being proposed, not a shipped cross-browser API. First model downloads take several minutes, and Opera says local inference performance is still being tuned.