LiteRT

At a glance

• Editor score
  8.8 / 10
• Pricing
  Open source
• Best for
  Cross-platform mobile and edge inference
• Paid from
  Free
• GPU acceleration
  Yes
• Facts checked
  13 Oct 2026

• Where it wins

  • CompiledModel supports CPU, GPU, and NPU inference
  • Quantization options include dynamic-range, integer, and float16
  • LiteRT.js enables browser inference with WebAssembly and WebGPU/WebNN

• Where it doesn't

  • Requires model conversion and deployment work rather than offering hosted inference
  • Legacy Interpreter and CompiledModel APIs may complicate runtime choices
  • Support channels are documentation and community

Our verdict on LiteRT

LiteRT is Google’s open-source runtime for running machine-learning and generative AI models on devices. It is aimed at developers who need inference across mobile, desktop, browser, and embedded environments, rather than a hosted inference service. The runtime supports TensorFlow, PyTorch, and JAX model deployment; PyTorch models can be converted to .tflite format. Its Apache-2.0 license and range of deployment targets make it relevant to solo developers as well as larger teams building on-device applications.

The runtime offers two paths for inference: the CompiledModel API for streamlined CPU, GPU, and NPU execution, and the legacy Interpreter API for backward-compatible inference. Hardware acceleration and post-training quantization are central capabilities; supported quantization methods include dynamic-range, integer, and float16. LiteRT.js extends inference to the browser using WebAssembly and WebGPU/WebNN, while LiteRT-LM supports generative AI inference. The C++ Tensor API is available for tensor and graph authoring. These options provide breadth, but teams should expect to make implementation choices around model conversion, runtime APIs, and target hardware.

LiteRT fits projects that prioritize local inference across platforms and can manage their own model and deployment workflow. It is less suited to buyers seeking a managed inference endpoint, paid service tiers, or direct vendor support channels: the project is open source, with documentation and community support listed. Choose it when control over on-device execution, acceleration, and quantization matters; consider a hosted alternative when minimizing infrastructure and runtime integration work is the priority.

Your product? Our editors can test LiteRT in depth and publish a review, labelled as paid · from $149/mo →

LiteRT pricing

LiteRT fact sheet

Alternatives to LiteRT

• NTNVIDIA TensorRTA free SDK for compiling and running optimized deep-learning inference on NVIDIA GPUs.9.4
• vLLMA flexible open-source engine for teams serving LLMs across varied infrastructure.9.3
• ATApache TVMA flexible open-source compiler for deploying models across CPUs, GPUs, mobile and edge hardware.9.2

See all LiteRT alternatives →

LiteRT vs the competition

• LNTLiteRT vs NVIDIA TensorRT
• LLiteRT vs vLLM
• LATLiteRT vs Apache TVM
• LLiteRT vs NVIDIA Triton Inference Server
• LORLiteRT vs ONNX Runtime
• LELiteRT vs ExecuTorch

Compare LiteRT with any tool side by side →

Featured on iTechGuides

LiteRT is listed in our AI Inference Engines directory. Add the badge to your site — it links back to this page.

<a href="https://www.itechguides.com/products/litert/"><img src="https://www.itechguides.com/best/badge/litert.svg" alt="Featured on iTechGuides" width="230" height="46"></a>

Reviewed by iTechGuides Editors · Editorial team · Updated Oct 2026

Advertiser disclosure: iTechGuides is reader-supported. Vendors can pay for top positions in our rankings and for a place on other products' pages, and we may earn a commission when you click some links. Bought places stand at the number bought; this page and our disclosure say so. Paying vendors may get a fuller, more detailed review from our editors, and that review can change their score. How we rank.

Last updated 29 Oct 2026 · How we research and update