Opera WebAI Early Preview: Local AI Models in the Browser

Introduction

On October 27, 2026, Opera (NASDAQ: OPRA) said it had begun shipping local AI models in its browsers. The first capability is WebAI, available as an early preview in Opera One, with Opera GX and Opera Neon to follow, according to the company's press release. Opera's developer blog describes the preview as spanning Opera One, Opera GX and Opera Neon, so Opera One is the safest reference point for what is available today.

WebAI lets websites and browser extensions use AI that runs on the user's own computer through the browser. Opera describes it as a proposed web standard backed by companies such as Google and Microsoft. Opera did not present the standard as its own invention; its contribution is an implementation that is model-agnostic and built around open models. This review looks at what that implementation offers, what it costs users in setup and hardware, and where the stated limits are.

Feature Overview

Three APIs for local models. WebAI introduces the Prompt, Decision and Semantic Embeddings APIs. In Opera's description, developers can generate text, classify information and find relationships between pieces of content. These are building blocks, and Opera frames them as the basis for tools that do more than answer questions in a chat window. The documentation is at next.opera.com/web-ai.

Models at launch. Opera says users can already use Google's Gemma and, on Macs, Apple's on-device foundation model. WebAI also works with compatible models already on the device, and open-weight models provided through Opera run on Opera's own inference engine. The aim, in Opera's words, is that developers are not tied to a single model provider.

Nothing extra to install. Local inference is built into the browser, so there is no separate AI app, no local server to configure and no API key to obtain. Because the AI runs on the user's machine, Opera says the content it processes does not need to be sent to a cloud AI provider. For developers, Opera lists no per-request cloud costs and no AI infrastructure to run.

Opt-in with permissions. Nothing is downloaded until the user turns WebAI on. Site-specific permissions let users decide which websites can access local AI, and the feature can be turned off at any time.

Example extensions. Opera released six experimental extensions: Project Lens, Got It?, Questionable Intelligence, Semantic Art, RecordIn Alchemy and C-I-A-O. Opera calls them "examples to spark ideas, not Opera features." They range from coloring an article by how much each paragraph matters to a project, to generating a quiz about an article just read.

Opera also launched HTML-in-Canvas alongside WebAI, which places live web pages inside 2D and 3D scenes, but it is a separate feature and not the subject here.

Usability Analysis

Opera's own try-it steps are short. Download or update Opera One, open Settings, then Opera AI (or opera://settings/ai), and turn on Enable Web AI API. Doing so downloads a compatible local model, which Opera says may take a few minutes. Then install one of the example extensions and open it from the extensions menu.

The platform support is narrower than the browser's reach.

Opera states that because the AI runs on the user's own device, performance depends on the hardware. Opera does not publish response-time figures, model sizes or minimum hardware requirements in the launch materials, so users should expect to test on their own machines.

The intended audience is plain in Opera's examples: independent developers and small teams who want a personal extension, for instance one that sorts tabs, flags clickbait or surfaces relevant job posts, without an account, a fee or a call to an AI service. Krystian Kolondra, EVP Browsers and AI at Opera, put the thesis this way: "A lot of what people want from AI doesn't need a data center. Your laptop can already do it, without sending your data anywhere."

Pros and Cons

Strengths. The built-in engine removes setup steps that usually deter casual experimentation. The opt-in design and per-site permissions give users control. Model-agnostic support, including models already on the device, reduces dependence on a single provider. For developers, the absence of per-request cloud costs is a concrete economic difference.

Limits. The early preview excludes several platforms, performance depends on the user's hardware, and WebAI is still only a proposed standard, so an extension built for it today relies on a feature other browsers may or may not implement. The press release also says Opera GX and Opera Neon are still to follow.

Outlook

Opera says this is the start of a broader push: more open models optimized for different devices and more tools for building browser experiences on local AI. It traces the effort to April 2024, when, by its own account, it was the first major browser to offer built-in access to local AI models in an experimental developer build. The company claims that today's launch brings that work into everyday browsers.

Whether WebAI matters beyond Opera depends on the proposed standard's progress, which the launch materials do not describe in detail. Watch for the arrival of GX and Neon support, wider platform coverage, published performance data and the model lineup.

Conclusion

WebAI is a sensible, clearly scoped early preview. It turns local models into something a browser extension can call without keys, servers or cloud billing, and it keeps the user in charge of downloads and permissions. It is best suited to extension developers, privacy-minded users on supported hardware, and anyone curious about on-device AI. Those on Intel Macs, Linux or Windows on Arm, or anyone who needs benchmark numbers first, will have to wait. Our rating of 3 reflects an early preview with promising design but unpublished performance details.

Editor's Verdict

Opera WebAI Early Preview: Local AI Models in the Browser is a workable proposition that fills a clear gap, even if it doesn't fundamentally change the landscape.

The strongest case for paying attention: a built-in inference engine removes the need for a separate app, a local server or an API key. That alone raises the bar for what readers should expect in this space. Reinforcing that, an opt-in design with site-specific permissions keeps the user in control of downloads and access — practical value rather than just headline appeal. The broader signal worth registering is straightforward: the notable step is a browser shipping a proposed web standard that Google and Microsoft are named as backing, not Opera introducing a new one. On the other side of the ledger, one constraint is real rather than a marketing footnote: the early preview supports only 64-bit Windows PCs and Apple-silicon Macs, so Intel Macs, Linux and Windows on Arm users are excluded. It should factor into any serious decision. Layered on top of that, a dependence on the user's own hardware means response speed varies by device, and the launch materials give no benchmark figures — which narrows the set of teams for whom this is an obvious yes.

For extension developers and privacy-minded users on supported Windows or Apple-silicon hardware, this is worth trying now; others can track it and revisit once platform coverage and performance details arrive.