Google released LiteRT.js, a JavaScript binding of LiteRT—Google’s on-device inference library previously called TensorFlow Lite—that runs .tflite models directly inside the browser via WebGPU, citing enhanced user privacy, zero server costs, and ultra-low latency because inference stays local. LiteRT.js is not a new model format; Google compiled its existing native runtime to WebAssembly and exposed it to JavaScript, shipping the native cross-platform runtime with its optimizations intact. Under that runtime, LiteRT.js targets three backends: CPU uses XNNPACK with multi-thread support and a relaxed SIMD build; GPU uses ML Drift through WebGPU; NPU uses the WebNN API, currently experimental in Chrome and Edge. LiteRT.js does not support partial delegation—a graph cannot split across CPU and GPU—and delegation is all-or-nothing per model, falling back to wasm execution if a model cannot be fully delegated. Google reports LiteRT.js is up to 3x faster across CPU and GPU inference against other web runtimes for classical computer vision and audio processing models, and GPU or NPU delivers a 5–60x speedup against its own CPU execution for demanding real-time work like object tracking and audio transcription, both benchmarks run in a controlled browser environment on a 2024 MacBook Pro with M4 Apple Silicon. Torch converts PyTorch models to .tflite in a single step, but the model must be export