Opera’s WebAI standard moves AI models into the browser

By Rebecca Stone

Opera has introduced WebAI, a web standard that runs AI models directly inside its browser without cloud processing, as an early preview in the flagship Opera One browser.

Opera says the design lets developers add AI features without running AI infrastructure or paying for every request. Users can keep information handled by a local model on their own device and avoid a cloud AI subscription.

The release matters because it targets two costs attached to cloud AI: server-side operating expense for developers and recurring subscriptions for users. Opera also presents local execution as a response to privacy concerns around assistants that receive personal information.

What does WebAI actually do?

Opera says anyone can perform three core tasks directly in the browser:

• Generate text.
• Classify content.
• Understand how pieces of information relate.

The company presents these as browser-native capabilities rather than features that require a hosted model service. It also says WebAI does not tie itself to one model. Opera designed the standard to work with a variety of AI models, including a wide range of open-weight systems.

Opera said in its announcement: “Developers can add AI features without running AI infrastructure or paying for every request.” The statement puts the economic case for WebAI in concrete terms: developers can add AI functions without shouldering the cost of every request.

Opera also said in the announcement: “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.” Opera is making both the developer economics and the user proposition central to the launch.

For developers, WebAI also opens a path to customization. Opera says the public availability of open-weight model weights gives developers freedom to adapt how a model works. That makes WebAI a way to deliver models developers can download and configure.

Which models and browsers will support it?

Opera names Google’s Gemma as one WebAI example and describes it as an open-weight model. The company says Google first introduced Gemma two years ago.

On Macs, WebAI also supports Apple’s on-device foundation model, according to Opera. Opera presents both as examples but says neither defines the standard’s scope.

The early rollout starts in Opera One, Opera’s flagship browser. Opera GX, built for gamers and streamers, and Opera Neon, which uses AI agents to perform tasks, will soon receive the same local-model treatment.

Opera says users must explicitly opt into the feature. The requirement makes local AI an explicit user choice rather than an automatic feature in the early preview.

WebAI grew out of Opera’s previous work to bring locally run models into its browsers. The new standard packages that effort under a broader, model-neutral name and extends the initial preview toward the company’s gaming and agent-focused products.

The expansion gives the standard three browser audiences without changing the core promise of local processing and explicit user choice. Opera still describes WebAI as one web standard rather than a separate implementation for each product.

Why does local execution matter for privacy?

Opera describes open-weight models as typically free to download and use. The company says developers can access their underlying weights publicly, which gives them more freedom to customize how the models work.

Opera draws a deliberate line between open-weight and fully open-source software. Open-weight models can still carry restrictions. The company therefore does not present public weights as equivalent to a fully open-source release.

Under WebAI, processing occurs on the user’s device rather than a company’s cloud server. Opera says users do not have to send personal information to company servers when they use local AI features. That arrangement gives users more control over information handled by the model.

Opera frames the privacy stakes around the growth of more capable AI assistants. The company says users often surrender vast amounts of personal information to AI chatbots, sometimes without a clear understanding of how that data is stored or used.

The local alternative does not eliminate every privacy risk, Opera acknowledges. It does, however, give users more control over processing on their own computers. That distinction matters because a privacy benefit tied to local execution is narrower than a claim that local AI is risk-free.

Opera has made digital privacy one of its primary selling points over the years. WebAI extends that position into AI assistants, which Opera says increasingly receive personal information and send it to company servers.

The commercial stakes are specific. Cloud AI products can charge for subscriptions and require infrastructure for each request. Opera says WebAI removes both requirements from the developer and user proposition, at the cost of relying on the hardware in each user’s hands.

Opera’s pitch is distinct for each group. For developers, the appeal is access to AI functions without a hosted endpoint. For users, it is access without a cloud subscription and with local handling of the information passed to the model.

What trade-offs remain?

Local execution places more demand on the user’s computer. Opera says that demand can affect performance. The company notes that model performance will depend on the device’s hardware.

Opera’s model-neutral design broadens model choice, but the company still ties performance to each device’s hardware. The same browser feature will therefore depend on the machine running it.

The early-preview label also matters. Opera is exposing the feature first in its flagship browser before bringing it to two distinct products. That staged approach keeps the initial test in one place while preserving a route to users who game, stream, or already use AI agents.

For users, the trade-off is concrete. They can avoid a subscription and keep local processing on their own device, but they need a computer capable of running the model and may see performance vary with its hardware.

For developers, Opera says WebAI lowers the infrastructure requirement and request cost for adding AI features. It also leaves performance tied to the hardware each user owns, the constraint Opera identifies.

WebAI’s privacy case rests on local processing and user control, not the disappearance of risk. Its performance case rests on the hardware under each model. Those conditions make the rollout broad in model support but dependent on each machine.

The next test is whether Opera can turn those trade-offs into consistent local AI performance as WebAI moves from Opera One to Opera GX and Opera Neon.