Skip to main content

Google's LiteRT.js Turbocharges AI in Your Browser, Up to 3x Faster

Google has officially released LiteRT.js, a new JavaScript library designed to supercharge AI inference directly in web browsers. The library leverages WebAssembly (Wasm) technology, combined with hardware acceleration APIs like WebGPU and WebNN, to replace the traditional JavaScript kernels used in TensorFlow.js. The result? AI and machine learning workloads in the browser can run up to three times faster.

Image

WebAssembly is a binary instruction format that lets web pages execute high-performance code at near-native speeds. TensorFlow.js, Google's earlier library, already allowed developers to run TensorFlow models in browsers. LiteRT.js is a major architectural upgrade on that foundation.

Tested 3x Speed Boost, Even on Older Hardware

Google tested LiteRT.js on a 2024 MacBook Pro with an M4 chip and saw inference speeds up to three times faster than the existing solution. However, the company notes that actual performance may vary on older hardware or different browser engines.

From a technical standpoint, Google is pushing web-based AI inference from the pure JavaScript era into a new phase that combines WebAssembly with hardware acceleration. As AI model speeds in browsers approach those of native apps, the experience of using AI features without installing a client—just by opening a web page—will become much smoother. This is a big win for the democratization of edge AI.

Key Points

  • LiteRT.js uses WebAssembly and hardware acceleration (WebGPU, WebNN) to speed up AI inference in browsers.
  • Up to 3x faster than TensorFlow.js on modern hardware, with variable performance on older devices.
  • No installation needed—AI runs directly in the browser, making edge AI more accessible.