Google Releases LiteRT.js: A JavaScript Binding of LiteRT That Runs .tflite Models in Browsers via WebGPU

Short summary

Google released LiteRT.js on July 9, 2026, a JavaScript binding for its LiteRT on-device inference library that executes .tflite models directly in browsers via WebAssembly and WebGPU. The runtime supports XNNPACK on CPU, ML Drift over WebGPU, and experimental WebNN for NPUs, with Google reporting up to 3x gains over competing web runtimes and 5–60x speedups for GPU or NPU paths versus CPU. A notable caveat is that tensors must be manually managed and explicitly deleted, adding complexity for developers adopting the library.

• •LiteRT.js runs .tflite models in-browser via WASM, WebGPU, and experimental WebNN
• •Google reports 3x over other web runtimes and 5–60x GPU/NPU gains over CPU
• •Tensors require manual memory management — developers must explicitly delete them

Generated with AI, which can make mistakes.