AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Opera introduces a new web standard to let users deploy AI locally in their browser

Collected Oct 27, 2026

Opera has announced WebAI, a new web standard it says allows AI models to run directly inside its browser instead of on cloud servers. The feature arrives first as an early preview in Opera One, the company's flagship browser, and users must opt in before it is active. Opera says its gaming- and streaming-focused browser, Opera GX, and Opera Neon, its browser built around AI agents executing tasks, will receive the same capability soon.

The standard is designed to be model-agnostic. Opera cites Google's Gemma, an open-weight model Google first released two years ago, as one example, and says WebAI also supports Apple's on-device foundation model on Macs. Beyond those two, the company positions WebAI as built for a broad range of open-weight AI systems rather than tied to a single vendor. The work builds on Opera's earlier efforts to bring locally run models into its browsers.

In terms of capability, Opera says WebAI lets anyone generate text, classify content, and reason about how pieces of information relate, all within the browser. Those are the canonical tasks for small language models: generation covers drafting and summarization, classification covers tagging and routing, and relationship understanding covers retrieval-style and semantic matching work.

Open-weight models are the technical foundation here. Their weights are publicly downloadable, and they are typically free to use, which gives developers room to customize behavior. They are not the same as fully open-source software, since open-weight releases usually carry license restrictions on how the weights and outputs may be used. This distinction matters for product teams evaluating whether a given model can be shipped commercially.

Opera's pitch rests on two arguments. The first is cost and infrastructure: in its announcement the company said developers can add AI features without running AI infrastructure or paying for every request. That framing targets the per-token billing model of hosted inference APIs, where costs scale with usage and engineering teams must provision capacity or manage rate limits. The second is privacy and control. Opera said users do not need AI subscriptions, data handled by local AI stays on their own device, and they remain in control of the AI on their computer.

Privacy has become a more visible concern as assistants grow more capable and are given access to personal information. Users frequently hand large amounts of personal data to chatbots with limited visibility into how it is stored or used. On-device processing offers an alternative because the computation never leaves the machine. It does not remove every risk, since the browser, extensions and the model itself still sit on the device, but it shifts the trust boundary. Opera has long marketed digital privacy as a selling point, which makes local inference a natural extension of that positioning.

The trade-offs are real. Local inference shifts compute cost onto the user's machine, which can affect performance, and Opera notes that model performance depends on the device's hardware. In practice this suggests meaningful variance across users: a recent laptop with a capable GPU or neural engine may run small models comfortably, while older or low-end hardware may see slow generation or limited model sizes. This is my own reading, but it follows from how on-device inference generally works, where memory bandwidth and available RAM constrain which models can load at all. It also implies a split in what developers can assume: an approach that works well on one user's machine may be unusable on another's.

Calling WebAI a web standard is a notable framing choice. A standard implies a specification that other browser vendors could implement, which would let a single web API reach users across Chrome, Safari, Firefox and others. Whether other vendors adopt it is unknown from Opera's announcement, and until they do, the practical reach of WebAI is limited to Opera's own user base. A likely trade-off is that early adopters get local AI features tied to one browser, while broader portability depends on the standard gaining traction elsewhere. The mention of Apple's on-device foundation model on Macs hints at a bridge to platform-provided models, which could reduce the download burden compared with shipping full model weights.

For developers, the appeal is a distribution path for AI features that avoids standing up inference infrastructure or paying per request, at the cost of unpredictable performance across devices and dependence on one browser's implementation. For product teams, it opens a privacy-forward option for sensitive use cases such as text classification or content analysis that users may not want routed through a third party.

Why it matters: developers building AI features now have a proposal for running models in the browser on the user's hardware, which could cut inference bills and keep data local, though results will vary by device and the standard's reach depends on adoption beyond Opera. Teams weighing it should test on representative low-end hardware, since performance is explicitly hardware-dependent. Product managers, meanwhile, get another way to market privacy without a subscription, and a reason to watch whether other browsers follow.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

The new standard, WebAI, is designed to work with a variety of AI models.