Opera announced an opt-in WebAI preview on October 27, 2026, starting with Opera One. It gives websites and browser extensions access to AI processing on the reader’s computer.
One example is Got It?, a separately installed extension that Opera says creates a quiz from the article you are reading. Its listing says the article and your answers stay on the device. Opera describes its launch extensions as experimental examples, not built-in browser features.
Treat the quiz as a reading aid, not a fact-check: generating questions from a page does not establish that the page itself is accurate.
Who can try it
Opera’s newsroom announcement says GX and Neon will follow Opera One, while its launch blog groups all three browsers together. Opera One is therefore the clearest starting point; simultaneous availability across all three is not established.
The supported computers are 64-bit Windows PCs with Intel or AMD processors, and Macs using Apple silicon. Intel Macs, Linux computers and Windows on Arm are excluded from the preview.
Opera’s setup instructions say to update or download Opera One, open Settings → Opera AI and switch on Enable Web AI API. This downloads a compatible model and may take several minutes. Install the quiz extension separately, then open it through the extensions menu.
Opera says response speed depends on the computer. Its launch material does not specify a minimum memory requirement or model download size. The quiz extension’s 148.4 KB package is not the size of the separate AI model or a measure of the memory needed to run it.
What changes when the browser runs the AI
Opera already offered experimental local-model chat in 2024. The change here is access for websites and extensions: Opera says they can generate text, sort information into categories and match related content. The browser handles the AI processing, without a separately configured model server or an inference API key.
For small teams building these tools, Opera says local processing avoids per-request cloud AI charges; users do not need an AI subscription for that processing. The tradeoff is that the reader’s hardware does the work. This does not make every connected website or service free, or establish that local AI will be faster than a cloud service.
Opera also says users can choose which websites may access local AI. Keeping model processing on the device removes the need to send that input to a cloud AI provider. It does not, by itself, establish where the rest of a website or extension sends information, or whether the whole tool works offline.
For privacy-sensitive reading, check the particular tool’s permissions and data handling, not just the location of its model. That distinction also applies to standalone local assistants: on-device computation and the behavior of connected services are separate questions.