Opera announced WebAI on October 27 as an early preview for websites and extensions using local AI models. It names Opera One, GX, and Neon. This concerns a proposed standard, rather than universal support.
What developers can build
The separate developer portal lists playgrounds for prompting, summarization, writing, rewriting, translation, proofreading, semantic embeddings, and decisions. The page includes checks for whether a particular API is available in the current browser. A card in the catalog does not establish that every function works on every device or in every browser configuration.
Experimental extensions shown there include highlighting paragraphs relevant to a project and asking questions about material the reader has just read. These are developer demonstrations. SIExpo has not installed them or measured their quality. The useful development is a more direct route from a web application to a local model that developers can examine and evaluate.
A local model changes the cost boundary
Our editorial reading is that avoiding a separate cloud request does not make processing free in an engineering sense. Work moves onto the user's device. A team must decide whether its task suits that environment and what behavior to offer when the required capability is unavailable. A responsive demonstration on one machine cannot establish performance for the entire intended audience.
Consider our example of semantic search over personal notes. Keeping the content local could be useful, but the application still needs to show whether its index is ready and includes recent changes. If a record is deleted, we would check whether it also disappears from search results. This is a proposed application test, rather than an observed Opera problem or a feature we have deployed.
Answer quality deserves a separate assessment. Passages with similar meaning can differ in date or in an important condition. Finding nearby text should therefore not count automatically as fulfilling the user's intention. In a pilot, we would preserve links to the original notes and examine incorrect matches as well as the successful examples.
Which devices are included
The preview covers Windows x64 on Intel/AMD and Apple-silicon Macs, excluding Linux, Intel Macs, and Windows on Arm. Opt-in downloads a model; speed is not guaranteed.
Local model inference alone does not establish that an entire website has no external connections. Our recommendation is to describe the model, synchronization, and other services separately, so a user can understand what stays on the device and what follows other settings. WebAI broadens the opportunities for building local features. The maturity of an individual product will depend on support, clear behavior when an API is missing, and evaluation against the real tasks its users need completed.
*AI illustration: a conceptual editorial image, not a documentary photograph or an actual product screen.*
*Editorial verification: 2026-10-28T17:14:21.370659+00:00.*
