Opera is testing a way to make the browser itself part of the AI stack. Its new WebAI standard is designed to let users run AI models locally, on their own devices, instead of sending every request to a cloud server. That could make AI more private and cheaper to operate. It could also make the computer doing the work much more important again.
Opera is presenting the standard as model-agnostic, with Google’s Gemma and Apple’s on-device foundation model cited as examples. The first implementation is available as an early preview in Opera One, with Opera GX and Opera Neon expected to follow. Users must opt in.
The browser becomes an AI runtime
The important detail is that WebAI is not simply another chatbot placed inside a browser tab. Opera says the standard can support tasks such as text generation, content classification, and understanding how pieces of information relate. In other words, it is being positioned as a way for developers to add AI behavior to browser experiences without building a separate cloud infrastructure for every request.
That changes the browser’s role. For years, browsers have mostly been the front door to remote services. The interface lived on the device, but the meaningful computation happened elsewhere. WebAI points toward a browser that can also provide access to local models, using the hardware already sitting underneath the screen.
The model choice matters here. Opera says WebAI is not tied to one proprietary system and is intended to work with a range of open-weight models. That gives developers more room to customize what they use, while potentially avoiding a per-request bill from an AI provider. It also makes the feature less dependent on one model company’s roadmap.
For developers, the appeal is obvious: add local AI capabilities without running AI infrastructure or paying for every inference. For users, the benefit is more tangible. Information handled by the local model can remain on the device, and some use cases may work without an AI subscription or a constant connection to a cloud service.
Local AI changes the trade-off, not the problem
The privacy argument is strong, but it is not magic. Keeping processing on a computer can reduce the amount of personal data sent to a company’s servers. It does not automatically make every AI feature accurate, secure, transparent, or well designed. Developers still have to explain what a model is doing and what information the surrounding browser experience can access.
There is also a basic hardware constraint. Local models need local resources. Opera says performance will depend on the user’s device, which means the same feature could feel fast on one computer and frustrating on another. The cloud has generally hidden that unevenness by concentrating the expensive work in a data center. Local AI puts the cost back in front of the user as processor demand, memory use, battery drain, or download size.
That tension will shape adoption more than the standard’s name. People may value privacy, but they will not necessarily accept a browser that slows down while trying to provide it. Developers may like avoiding inference costs, but they will still need to design around different hardware profiles and model limitations.
Opera’s move is therefore less about replacing cloud AI than creating another place for AI to live. The browser can become a control point where users choose between speed, privacy, cost, and capability depending on the task. That is a meaningful shift for the web, because it moves AI from being only a remote service into something that can be delivered as part of the user’s everyday software environment.
The strategic consequence is clear: if local models become reliable enough, the browser will no longer be just where people access AI products. It may become the layer that decides which AI runs locally, what data leaves the device, and how much of the experience belongs to the user rather than the service provider.