Opera has added WebAI to its Opera One browser, an opt-in early preview that lets websites and browser extensions use an AI model running on your own computer. The model is Google’s Gemma, or Apple’s on-device model on a Mac. It can write text, sort content and spot related passages without a cloud AI service, an account or an API key. Opera announced the preview on 27 October 2026 for 64-bit Windows PCs and Macs with Apple silicon.
One switch in Opera’s AI settings downloads the model, and each site needs your permission
WebAI is off by default. To try it, update Opera One, open Settings, go to Opera AI and turn on “Enable Web AI API”. Opera says this downloads a local model, which can take a few minutes. Nothing is downloaded before you flip the switch, you can turn it off at any time, and a website needs your permission before it can use the model.
Opera also published six experimental extensions to show the idea. Project Lens, for example, highlights the paragraphs of an article that matter for your project. Developers can try the APIs directly in Opera’s WebAI playground.
- Works on: Opera One on 64-bit Windows PCs with Intel or AMD chips, and on Macs with Apple silicon.
- Not yet on: Intel Macs, Linux and Windows on Arm. Opera GX and Opera Neon follow later, without a date.
- Models: Google’s Gemma, Apple’s on-device model on Macs, and compatible models already on your computer.
- Price: free, with no subscription.
Opera does not say how large the download is or how much memory the model needs, only that speed depends on your hardware. As a reference point, Chrome’s own local model, Gemini Nano, takes about 4.27 GB, Tech Times reported.
“A lot of what people want from AI doesn’t need a data center. Your laptop can already do it,” said Krystian Kolondra, Opera’s EVP Browsers and AI.
Opera calls WebAI a proposed standard, but Safari and Firefox are not on board
For developers, the draw is cost. A site or extension can summarise, rewrite or classify text on the visitor’s machine, so there is no AI server to run and no bill per request. Opera’s developer pages list eight APIs, among them the Prompt, Summarizer and Translator APIs that Google also offers in Chrome.
Opera says Google and Microsoft back the proposal, and both have built similar features. Google switched on its Prompt API, which runs Gemini Nano on the user’s computer, in the regular release of Chrome, version 148, on 5 May. Microsoft has offered the Prompt API in test versions of Edge since 2025, first with its Phi-4-mini model.
That does not make it a settled standard. When Chrome shipped the Prompt API, Mozilla and Apple’s WebKit team formally opposed it, and Microsoft and the W3C’s Technical Architecture Group raised concerns, Tech Times reported. The objections cover privacy, battery drain and interoperability: the same website can behave differently depending on the model each browser ships. Every Safari and Firefox user still needs a fallback, such as a cloud model or no AI feature at all.
Does local AI in the browser keep your data on your device?
Partly. The AI step runs on your laptop, so the text it handles does not go to a cloud AI provider. But the website or extension that calls the model already sees your page, and it can still send the results to its own server, like any other site. Local AI removes one party from the chain, not all of them. Opera’s own wording is careful too: it says page data “can stay” on your device.
Gemma on a laptop can sort documents, but Opera’s preview is no base for company tools
Plenty of organisations hold back from cloud AI because of the GDPR or client contracts. For them, WebAI shows where things are heading. Small open models such as Gemma now run on an ordinary work laptop. They can handle narrow jobs, such as tagging incoming email or sorting documents, without the text leaving the device.
A browser preview is still the wrong foundation for a business process. Opera picks the model, not you, and it says nothing about settings for IT administrators. Small models also make more mistakes than large cloud models on long or complex documents. A local model on your own server or laptops gives the same privacy gain, with a model and version you choose.
Our advice: pick one narrow task that is now stuck on privacy, such as classifying incoming documents, and run it on a small model like Gemma. Count how often it gets the answer right compared with your current process. If the quality holds, build on a local model you control, and treat WebAI as a sign that browsers will offer this route later.
For more information, visit the official announcement of WebAI on the Opera blog.
