Opera unveils WebAI, a standard for running AI models right in the browser

Opera has introduced a new web standard called WebAI. It allows AI models to run directly in the browser, without depending on cloud services.

WebAI is designed to work with a variety of models. As one example, Opera points to Gemma, the open-weight model Google first introduced two years ago. On Macs, the standard also supports Apple's on-device foundation model. The company stresses, however, that WebAI isn't tied to any single model and is built to support a wide range of open-weight systems. The standard grew out of Opera's earlier work on bringing locally run models into its browsers.

Where and how to try it

The new capabilities are available as an early preview in Opera One, the company's flagship browser. Opera GX, its browser for gamers and streamers, and Opera Neon, which uses AI agents to perform tasks, will get them soon. The feature is opt-in.

According to the company, WebAI can be used to generate text, classify content and understand how pieces of information relate to each other — all inside the browser.

Open weights and privacy

Open-weight models are typically free to download and use, and their publicly available weights give developers more freedom to customize them. Unlike fully open-source software, though, they still come with some restrictions.

Opera also highlights privacy: users don't have to send personal data to a company's servers or pay for a cloud subscription. "Developers can add AI features without running AI infrastructure or paying for every request," the announcement says. "Users don't need AI subscriptions, the information the local AI handles stays on their own device, and they stay in control of the AI on their computer."

Privacy is becoming a sharper concern as AI assistants gain access to personal information. Users often hand chatbots vast amounts of data without fully understanding how it is stored or used. Local processing doesn't remove every risk, but it gives users more control. It's no surprise that Opera, which has long pitched privacy as one of its main selling points, chose this route.

There is a downside: running models locally puts more strain on the computer and can affect performance. Opera acknowledges that model performance will depend on the device's hardware.

Source: TechCrunch