Opera previews WebAI, a proposed standard for running AI models locally in the browser

Centre · 1Opera has introduced WebAI, a new web standard that lets users run AI models directly in its browser without relying on the cloud. 1 Centre · 2It is available as an opt-in early preview in the flagship Opera One browser, and Opera says Opera GX and Opera Neon will soon get the same treatment. 2 1 Centre · 1Computerworld describes WebAI as a proposed standard that lets websites and browser extensions access locally running models for tasks such as generating text and classifying content. 2

Centre · 2Opera says WebAI is not tied to a particular model and supports a range of open-weight systems, naming Google's Gemma and, on Macs, Apple's on-device foundation model. 1 2 Centre · 1The preview supports 64-bit Windows PCs with Intel or AMD processors and Macs with Apple silicon, but not Windows on Arm. Opera does not specify a requirement for a neural processing unit and says performance depends on the device's hardware. 2 Centre · 1Models are downloaded only after users enable the feature, and website-specific permissions let users control which sites can access the local models. Opera plans to add more models optimised for different devices. 2

Centre · 1Opera argues developers can add AI features without running AI infrastructure or paying per request, and that users need no AI subscriptions and keep data on their own device. TechCrunch notes that local processing does not remove every privacy risk and puts more demand on the user's computer. 1 Centre · 1Computerworld quotes analysts on enterprise effects. Pareekh Jain of Pareekh Consulting said local processing could cut the need to send sensitive data to outside AI providers but could make device activity harder for IT departments to monitor, and warned that malicious websites could manipulate local models through hidden prompts and rogue extensions could reach sensitive on-screen information. 2 Centre · 1Tulika Sheel of Kadence International said enterprises still need controls over which sites and extensions can access local models, and that lower usage costs do not automatically mean lower total costs. Jain said smaller models could handle tasks such as summarisation on existing PCs, reducing the need for immediate hardware upgrades, while complex workloads will still need dedicated hardware or the cloud. 2

Centre · 1Whether WebAI has effects beyond Opera depends on wider adoption of the proposal as a standard. 2