Every developer who ships an AI feature eventually runs into the same bill. Each request costs money, and the cost grows with users. Opera thinks it has a way out, and it involves moving the model onto the user’s machine.
The company introduced WebAI, a proposed web standard for running AI models locally inside the browser. It isn’t built around one model. Opera names Google’s Gemma, the open-weight model Google released two years ago, as an example, and says that on Macs it also supports Apple’s on-device foundation model. The pitch is that sites and extensions can generate text, classify content, and work out how pieces of information relate, with no round trip to a server.
It’s an early preview in Opera One, and users have to opt in. Opera GX and Opera Neon, the company’s agent-driven browser, are expected to follow.
The appeal splits two ways. For developers, Opera says they can add AI features without running infrastructure or paying per request. For users, there’s no subscription, and the data the model handles stays on the device. Open-weight models help here because they’re typically free to download and their weights are public. They still carry license restrictions, so they aren’t the same as true open source.
The bigger picture is a quiet race to make the browser an AI runtime. Google has been pushing built-in AI in Chrome, including Gemini Nano, and Apple has its own on-device model. Opera’s angle is neutrality. If WebAI really works across models, it competes less with those efforts and more with the idea that every browser vendor ships its own incompatible API.
That’s also where I’d stay skeptical. A web standard only matters if other browsers implement it. Opera has a small market share compared with Chrome, Safari, and Edge, and developers rarely build for a feature that works in one browser. Until Chromium or Firefox signals interest, this is a nice Opera feature, not a standard.
Privacy is the strongest argument, and Opera has long made it a selling point. Local processing doesn’t remove every risk, but it keeps personal data off someone else’s server. That matters as chatbots absorb more sensitive information.
The trade-off is hardware. Local models put real load on a machine, and Opera admits performance depends on the device. A recent laptop will cope. An older one may crawl.
So is this noise? Not quite. The direction is right, and the cost argument for developers is real. But watch for a second browser to sign on. That’s the signal that matters.
