Opera launches WebAI standard for running AI models locally in browsers

Opera has introduced WebAI, a new web standard for running AI models directly inside the browser instead of on a remote server. The company announced the standard this week and is rolling it out first as an early preview in Opera One, its main browser.

WebAI is designed to work with open-weight models, whose underlying training weights are public, rather than being locked to one AI system. Opera points to Google’s Gemma, an open-weight model the company released two years ago, as one example that runs through WebAI. On Macs, the standard also taps Apple’s on-device foundation model, giving users another local option depending on their hardware.

Opt-in, and spreading to Opera’s other browsers

Users have to turn the feature on themselves; it isn’t enabled by default. Opera says the same local AI capability will soon extend to Opera GX, its browser aimed at gamers and streamers, and Opera Neon, which uses AI agents to carry out tasks on a user’s behalf. For now, Opera One is the only browser where the preview is live.

Opera frames the benefits in both cost and privacy terms. “Developers can add AI features without running AI infrastructure or paying for every request,” the company said in its announcement. Because the model runs on the user’s own machine, Opera says people don’t need a paid AI subscription, and the information the model processes never leaves their device. Through WebAI, Opera says the browser can generate text, classify content, and work out how different pieces of information relate to each other, all without a cloud connection.

That pitch lands at a moment when AI assistants are collecting more personal information than before, often without users fully understanding how it’s stored or reused. Running a model locally doesn’t remove every privacy risk tied to AI, but it does keep the raw data processing off a company’s servers, which is the kind of distinction Opera has leaned on for years as part of its broader privacy marketing.

The trade-off is performance. Running an AI model on a laptop or desktop instead of a data center demands more from that device’s processor and memory, and Opera acknowledges that how well WebAI performs will depend on the hardware it’s running on. A model that responds quickly on a high-end Mac may lag on an older or less powerful machine.

Opera hasn’t said which specific open-weight models beyond Gemma and Apple’s on-device system will be supported, nor given a firm date for when Opera GX and Opera Neon will get the feature. Those details, along with how WebAI performs across a range of everyday laptops, are what will determine whether the standard moves beyond an opt-in preview into something most Opera users actually turn on.