Opera Brings On-Device AI to the Browser With WebAI
Opera has introduced WebAI, a web standard that lets pages run AI models directly in the browser on the user's own machine, without sending the work to a cloud server. TechCrunch reported the launch on October 27, 2026.
The standard is designed to work with several models. Opera has named Google's Gemma, an open-weight model whose trained weights are published for anyone to download, as one of them. On Macs, WebAI also supports Apple's on-device foundation model. Through WebAI, a page can generate text, classify content and analyse how pieces of information relate to one another, all within the browser. Opera says performance depends on the hardware of the device running the model, so inference speed (how quickly the model produces results) and output quality will vary from machine to machine.
For now, WebAI is an early preview inside Opera One, and users must opt in to enable it. Opera plans to bring the feature to Opera GX and Opera Neon but has not given a timetable. A dependency is already visible in the extension ecosystem: Opera's Semantic Art extension, listed on Opera's add-ons site, requires Opera with built-in AI turned on.
The announcement builds on several years of AI work at Opera. Aria is the browser's native AI, and Opera has promoted local large language model support through its AI feature drops since 2024. In March 2025 the company said it had become the first major browser with AI-based agentic browsing, a mode in which natural-language instructions are converted into browsing tasks. In September 2025 it launched Neon, an AI-powered browser that can execute tasks and run code inside web pages. Opera had also said it expected to launch a Browser Operator as part of the same feature drop program.
In my view, the more consequential detail is architectural. Most browser AI so far has meant sending a request off the device to a hosted model, with the latency, cost and data-handling questions that follow. A standard that exposes local inference to web content moves part of that work onto the user's machine, which changes what a web developer can assume about that machine. It also raises practical questions: capability detection (how a page checks what a device can run), model download size and storage, and how a page should degrade when the local model is missing or too slow on older hardware.
The broader context is the split between on-device and cloud AI. Apple's foundation model and open-weight models such as Gemma give developers a local inference target that is not tied to a single vendor's servers, something a standard browser context did not offer a few years ago. Whether WebAI becomes a lasting part of the web platform will depend on whether other browser makers implement comparable interfaces.
The practical upside is that summarising, classifying and relating content on the page a user is already reading can run without a round trip to a server. For a privacy-conscious user, or for an application that cannot assume connectivity, that moves computation to a different place. The real test is how much of this capability survives outside a preview flag, and how well it performs on the mid-range laptops most people actually use.
