--- title: LiteRT.js, Google's high performance Web AI Inference description: Google has announced LiteRT.js, a JavaScript binding of its LiteRT on-device inference library that enables running ML models directly in the browser. It... canonical: https://daily.dev/posts/litert-js-google-s-high-performance-web-ai-inference-trneihldh twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: LiteRT.js, Google's high performance Web AI Inference | daily.dev og:description: Google has announced LiteRT.js, a JavaScript binding of its LiteRT on-device inference library that enables running ML models directly in the browser. It... og:url: https://daily.dev/posts/litert-js-google-s-high-performance-web-ai-inference-trneihldh og:image: https://api.daily.dev/og/posts/TRneIHlDh.png og:image:alt: LiteRT.js, Google's high performance Web AI Inference og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # LiteRT.js, Google's high performance Web AI Inference **[Google Developers](https://daily.dev/sources/googledevs)** · 5 min read · 3 upvotes · 0 comments ## Summary Google has announced LiteRT.js, a JavaScript binding of its LiteRT on-device inference library that enables running ML models directly in the browser. It supports hardware acceleration via WebGPU, XNNPACK (CPU), and WebNN (NPU), delivering up to 3x speedups over existing web runtimes like TensorFlow.js and 5-60x gains when using GPU/NPU over CPU. The library supports .tflite models, PyTorch model conversion via LiteRT Torch, and tailored quantization through AI Edge Quantizer. An npm package is available, and Ultralytics has integrated official LiteRT export support for YOLO models. Use cases include object detection, depth estimation, image upscaling, and audio processing entirely client-side with no server costs. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is LiteRT.js and how does it compare to TensorFlow.js for running AI models in the browser? LiteRT.js is a JavaScript binding of Google's LiteRT on-device inference library that runs .tflite models directly in the browser. Unlike TensorFlow.js, which relies on less performant JavaScript-based kernels, LiteRT.js exposes the native cross-platform runtime through WebAssembly, using XNNPACK for CPU and ML Drift for GPU, delivering up to 3x speedups over other web runtimes. _daily.dev surfaces runtime comparisons like this for developers choosing a client-side inference stack._ ### How do I convert a PyTorch model to run in the browser with LiteRT.js? PyTorch models can be converted in a single step using LiteRT Torch (ai-edge-torch), producing a format ready for browser-based hardware acceleration. The AI Edge Quantizer tool then lets developers configure tailored quantization schemes per layer to reduce model size and improve performance while preserving quality, with a getting-started guide and a selective quantization colab available. _developers converting PyTorch models for the web track tooling like this via daily.dev._ ### How much faster is GPU or NPU inference than CPU inference for browser-based AI with LiteRT.js? Using WebGPU or WebNN (via Apple CoreML) for GPU or NPU acceleration delivers 5-60x speedup compared to standard CPU execution via XNNPACK, based on benchmarks across object tracking, audio transcription, and image manipulation models. This makes real-time client-side inference tasks feasible without server round-trips. _daily.dev helps developers weigh backend acceleration options like this when planning real-time AI features._ ## Similar posts on daily.dev - [Google Launches AI Inference “LiteRT.js“](https://daily.dev/posts/google-launches-ai-inference-litert-js--golcrppmb) · C\# Corner · 2 upvotes · 1 comments - [LiteRT: The Universal Framework for On-Device AI](https://daily.dev/posts/litert-the-universal-framework-for-on-device-ai-dkvzmc8hv) · Google Developers · 0 upvotes · 0 comments - [Google’s LiteRT adds advanced hardware acceleration](https://daily.dev/posts/google-s-litert-adds-advanced-hardware-acceleration-negvjtwmm) · InfoWorld · 0 upvotes · 0 comments - [How Google Is Shifting AI From the Cloud to Your Browser](https://daily.dev/posts/how-google-is-shifting-ai-from-the-cloud-to-your-browser-gdtsilce3) · The New Stack · 1 upvotes · 0 comments - [WebGPU Browser AI: Run LLMs Client-Side, No Backend](https://daily.dev/posts/webgpu-browser-ai-run-llms-client-side-no-backend-1evlxlfvf) · SitePoint · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#webassembly](https://daily.dev/tags/webassembly), [#webgpu](https://daily.dev/tags/webgpu) [View this post on daily.dev](https://daily.dev/posts/litert-js-google-s-high-performance-web-ai-inference-trneihldh) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"LiteRT.js, Google's high performance Web AI Inference","url":"https://daily.dev/posts/litert-js-google-s-high-performance-web-ai-inference-trneihldh","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/litert-js-google-s-high-performance-web-ai-inference-trneihldh"},"datePublished":"2026-07-09T20:54:25.292Z","dateModified":"2026-09-14T09:08:16.378Z","description":"Google has announced LiteRT.js, a JavaScript binding of its LiteRT on-device inference library that enables running ML models directly in the browser. It...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/3b990d7ed3f7916acab017dc0b34c728?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/3b990d7ed3f7916acab017dc0b34c728?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Google Developers","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Google Developers","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/ea4bbded4e7f45ccb82e121a3156535f","url":"https://daily.dev/sources/googledevs"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/litert-js-google-s-high-performance-web-ai-inference-trneihldh","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,webassembly,webgpu","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Google Developers","item":"https://daily.dev/sources/googledevs"},{"@type":"ListItem","position":3,"name":"LiteRT.js, Google's high performance Web AI Inference"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/litert-js-google-s-high-performance-web-ai-inference-trneihldh#faq","mainEntity":[{"@type":"Question","name":"What is LiteRT.js and how does it compare to TensorFlow.js for running AI models in the browser?","acceptedAnswer":{"@type":"Answer","text":"LiteRT.js is a JavaScript binding of Google's LiteRT on-device inference library that runs .tflite models directly in the browser. Unlike TensorFlow.js, which relies on less performant JavaScript-based kernels, LiteRT.js exposes the native cross-platform runtime through WebAssembly, using XNNPACK for CPU and ML Drift for GPU, delivering up to 3x speedups over other web runtimes. daily.dev surfaces runtime comparisons like this for developers choosing a client-side inference stack."}},{"@type":"Question","name":"How do I convert a PyTorch model to run in the browser with LiteRT.js?","acceptedAnswer":{"@type":"Answer","text":"PyTorch models can be converted in a single step using LiteRT Torch (ai-edge-torch), producing a format ready for browser-based hardware acceleration. The AI Edge Quantizer tool then lets developers configure tailored quantization schemes per layer to reduce model size and improve performance while preserving quality, with a getting-started guide and a selective quantization colab available. developers converting PyTorch models for the web track tooling like this via daily.dev."}},{"@type":"Question","name":"How much faster is GPU or NPU inference than CPU inference for browser-based AI with LiteRT.js?","acceptedAnswer":{"@type":"Answer","text":"Using WebGPU or WebNN (via Apple CoreML) for GPU or NPU acceleration delivers 5-60x speedup compared to standard CPU execution via XNNPACK, based on benchmarks across object tracking, audio transcription, and image manipulation models. This makes real-time client-side inference tasks feasible without server round-trips. daily.dev helps developers weigh backend acceleration options like this when planning real-time AI features."}}]} ```