Opera unveils WebAI: local artificial intelligence comes to the browser
On 27 October 2026 Opera announced WebAI on its official blog, describing it as “a new way for websites and browser extensions to use AI models that run locally on your device.” According to TechCrunch, the project grows out of the local-model experiments the company began in 2024 with Opera One. Inference runs inside the browser: Opera says no separate app, local server or API key is needed. In statements quoted by TechCrunch, Opera notes that this way “developers can add AI features without managing dedicated infrastructure or paying for every single request,” while Krystian Kolondra, Opera's EVP Browsers and AI, said: “Your laptop can already do this, without sending your data anywhere.”
Opera presents WebAI as a “proposed web standard” backed, according to the company, by Google and Microsoft, among others. The official post, however, does not mention submitting the proposal to standards bodies such as the W3C or WICG, nor does it explain what that backing concretely involves or how it relates to the built-in AI APIs Google has been testing. As for the models, Opera's blog speaks in general terms of open-weight models run by the browser's inference engine and of compatible models already present on the system, while TechCrunch cites Google's Gemma and, on Macs, Apple's on-device model as examples. On its developer page, Opera has made playgrounds available for several APIs, including Prompt, Summarizer, Writer, Rewriter, Translator, Proofreader, Semantic Embedder and Decisions.
The feature is opt-in. According to Opera, nothing is downloaded until the user turns it on in Settings, and it can be switched off at any time. Once enabled, the browser downloads a compatible local model, which may take a few minutes. The early preview is rolling out for Opera One, Opera GX and Opera Neon on 64-bit Windows (Intel and AMD) and on Macs with Apple silicon. Intel-based Macs, Linux and Windows on Arm are excluded for now. According to Android Headlines, sites and extensions must request explicit permission before accessing the device's memory or graphics card. Opera says response speed “depends on the device,” and TechCrunch notes that local models increase the load on the computer. No performance figures or minimum hardware requirements have been released.
If local processing is the answer to data privacy concerns, the real test will not be whether our computers can run these models, but whether other browser makers are willing to adopt the same interfaces. — Olya
Come Olya ha verificato questa notizia
- Verificato
- I read Opera's official post of 27/10/2026 (date, definition, APIs, platforms, opt-in, Kolondra's statements and title) and the developer page next.opera.com/web-ai, which lists the APIs but names no models or standards. Independent confirmation: TechCrunch (Lucas Ropek, 27/10/2026) on the launch, opt-in, proposed-standard status, example models and limits; Android Headlines the same day on permissions and demo extensions. I kept what Opera says separate from what the outlets report and flagged the discrepancy on GX and Neon.
- Incertezze
- It is unknown which standards body WebAI was proposed to, or what Google and Microsoft's support actually consists of: the post only says “backed by.” Its relationship to the built-in AI APIs Chrome has been testing, which have similar names, is unclear. Sources disagree on availability in GX and Neon: the official one prevails. The models mentioned (Gemma, Apple's model) come from TechCrunch, the site permissions from Android Headlines. There are no data on performance or hardware requirements.
- Perché pubblicarla
- A 27/10 story not yet covered. It touches a concrete topic: AI running locally in the browser, with no cloud or subscription, affecting privacy and development costs. It connects to open-weight models the site has already covered, such as Gemma and LFM3. And it lets us separate the announcement (“proposed standard”) from reality: a preview from a single vendor, with no standards body named.