Opera 发布 WebAI，在浏览器中内置本地 AI 模型

Opera announced on October 27, 2026, that it began shipping local AI models directly in its browsers, making AI that runs on a user’s own computer a central part of the next generation of Opera browsers. The first capabilities are available as an opt-in early preview in Opera One, with Opera GX and Opera Neon to follow.

What WebAI Includes at Launch

WebAI is a proposed web standard that, according to Opera, is backed by companies such as Google and Microsoft and lets websites and browser extensions use AI hosted through the browser. Opera describes its implementation as model-agnostic and built around open models, offering users a choice among AI providers. At launch, users can tap models from Google and Apple, including Google’s Gemma and, on Macs, Apple’s on-device foundation model, directly in Opera. Opera says WebAI lets anyone generate text, classify content, and understand how pieces of information relate, all within the browser.

Krystian Kolondra, Opera’s EVP Browsers and AI, framed the release around what consumer hardware can already handle. “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,” he said. Kolondra said Opera first shipped local models to developers in 2024 and is now bringing them to everyone, adding that more models and tools are coming and that users will always decide what runs on their machine.

That earlier work began with an April 3, 2024 announcement, when Opera added experimental support for 150 local LLM variants from approximately 50 model families to the Opera One developer stream, including Meta’s Llama, Vicuna, Google’s Gemma, and Mistral AI’s Mixtral. Each variant typically required 2 to 10 GB of local storage, and the local models complemented Aria, Opera’s online browser AI service. Opera describes the 2024 developer build as making it the first major browser with built-in access to local AI models.

Controls, APIs, and Example Extensions

WebAI is opt-in: nothing is downloaded until the user turns it on in Settings, it can be turned off at any time, and site-specific permissions let users decide which websites can access local AI. According to Opera’s developer blog post, local inference is built into the browser, so there is no separate AI app to install, no local server to configure, and no API key to obtain. WebAI works with open-weight models provided through Opera’s inference engine and with compatible models already on the user’s device, so users are not tied to a single model provider.

Opera says the design means users do not need AI subscriptions and that information the local AI handles stays on their device, while developers can add AI features without running AI infrastructure or paying per request. The company gives the example of an extension that could sort tabs, flag clickbait, or spot relevant job posts using a model on the user’s laptop, with no account, no fee, and no data sent to an AI service.

WebAI introduces the Prompt, Decision and Semantic Embeddings APIs, which Opera describes as building blocks for generating text, classifying information, and finding relationships between pieces of content. Opera’s WebAI developer page also lists Summarizer, Writer, Rewriter, Translator and Proofreader APIs, with playgrounds that run entirely on-device in the browser.

To demonstrate what developers can build, Opera released six experimental example extensions it describes as idea starters rather than Opera features: Project Lens, which colors an article’s paragraphs by how much each matters to a stated project; Got It?, which writes a quick quiz about an article just read; Questionable Intelligence, which asks playful questions about a page and draws the verdict as a chart; Semantic Art, which turns an article into moving art with similar sentences becoming visual neighbors; RecordIn Alchemy, which turns any article into what Opera calls an insufferably inspirational social media post; and C·I·A·O, in which two ASCII cats improvise a scene on any topic.

Availability and Platform Limits

WebAI is available as an opt-in early preview on 64-bit Windows PCs with Intel or AMD processors and on Macs with Apple silicon. It is not available on Intel-based Macs, Linux, or Windows on Arm. Enabling WebAI in Settings downloads a compatible local model, a step Opera says may take a few minutes. Because the AI runs on the user’s own device, Opera says performance depends on the hardware and that it will keep working to make local AI lighter and faster. The company describes the launch as the start of a broader push and says it plans to bring more open models optimized for different devices, along with more tools for building browser experiences on local AI.

HTML-In-Canvas Launches Alongside

Also on October 27, 2026, Opera launched HTML-in-Canvas, a web capability that places real, interactive web pages inside 2D and 3D scenes. Opera says canvas content was previously rendered only as pixels that could not be clicked, typed into, or selected; with HTML-in-Canvas, menus, buttons, and forms can move with a scene and still work, built with the HTML and CSS developers already know. Opera presents the capability as aimed at game interfaces such as shops, inventories, settings, and chat, and at Opera GX, its browser for gamers. The company says it helped bring Canvas to the web in 2006 and later worked on WebGL.

“Honestly, this is a fun one,” said Maciej Kocemba, Product Director of Opera GX. “You can take a real web page and throw it into a 3D world, and everything still works: buttons, text, menus, all of it. We wanted users to get their hands on it early and see what they come up with.”

Opera’s demos include Webpage Bulldozer, in which a bulldozer pushes a real web page’s text, images, and buttons into piles; Gamer Room, a 3D gaming room whose screens are live, interactive web pages; and HTML Ninja, plus eight effects demos. According to Opera’s HTML-in-Canvas blog post, the capability is an emerging web platform proposal developed in the open and implemented in Chromium; same-origin rules apply, so developers can draw their own content into a canvas but not other websites. An Early Bird setting enables HTML-in-Canvas across the web, and the demos work in Opera browsers with no setup.