Opera launches WebAI, a web standard for running AI models locally in the browser

Opera has unveiled WebAI, a new web standard that lets users run artificial intelligence models directly in the browser, without relying on the cloud. According to TechCrunch, the standard is designed to work with different AI models and is not tied to any one in particular, as it aims to support a wide range of open-weight systems.

As an example, Opera cites Gemma, the open-weight model Google introduced two years ago. On Macs, WebAI also supports the foundation model that Apple runs on-device, according to the company.

WebAI grew out of Opera's earlier work to integrate locally run models into its browsers. The new features are available as an early preview in Opera One, the company's flagship browser. Opera says that Opera GX, its browser for gamers and streamers, and Opera Neon, its browser that uses AI agents to carry out tasks, will soon get the same treatment. Users must opt in to the feature.

The article explains the advantages of open-weight models: they are usually free to download and use, and their weights are public, giving developers more freedom to customise them. Unlike open-source software in the strict sense, these models retain some restrictions.

Opera also highlights privacy: users can use AI features without sending their personal data to a company's servers or paying for a subscription to a cloud AI service. In the company's words, developers can add AI features without running their own infrastructure or paying per request, and the information processed by the local AI stays on the user's device. WebAI can generate text, classify content and understand the relationships between pieces of information, all within the browser.

The piece frames the initiative within growing privacy concerns as AI assistants become more powerful and access more personal information. On-device processing does not eliminate all risks, but it gives users somewhat more control. Privacy has been one of Opera's main selling points for years.

There are trade-offs: running models locally places greater demands on the user's machine and can affect performance. Opera acknowledges that model performance will depend on the device's hardware.

Note: the content received is the TechCrunch article (by Lucas Ropek); it does not include technical details of the specification or performance figures, so this summary is limited to what was published.