Installation - ONNX Runtime

ONNX Runtime supports multiple platforms and programming languages. Choose your preferred language and platform below.

Python

ONNX Runtime for Python is available on PyPI for Windows, Linux, and macOS.

CPU
GPU (CUDA)
Other Execution Providers

Install the CPU-only version:

pip install onnxruntime

For nightly builds:

pip install onnxruntime --pre

Install the GPU version with CUDA support:

pip install onnxruntime-gpu

CUDA 12.x is required. The package automatically installs CUDA dependencies when you install with the [cuda] extra:

pip install onnxruntime-gpu[cuda]

For cuDNN support:

pip install onnxruntime-gpu[cuda,cudnn]

Additional execution provider packages:

# OpenVINO
pip install onnxruntime-openvino

# DirectML (Windows)
pip install onnxruntime-directml

# TensorRT
pip install onnxruntime-gpu  # TensorRT EP included

# ROCm (AMD GPU)
pip install onnxruntime-rocm

Requirements

Python 3.11 or later (3.11, 3.12, 3.13, 3.14 supported)
Compatible with Windows, Linux, and macOS

Optional Dependencies

# For symbolic shape inference
pip install onnxruntime[symbolic]

# For quantization utilities
pip install onnxruntime[quantization]

C/C++

ONNX Runtime provides pre-built C/C++ libraries for multiple platforms.

Download the Release Package

Download the appropriate package from GitHub Releases:

Windows: onnxruntime-win-x64-[version].zip
Linux: onnxruntime-linux-x64-[version].tgz
macOS: onnxruntime-osx-[arch]-[version].tgz
Mobile: onnxruntime-android-[version].aar, onnxruntime-ios-[version].xcframework

Extract the archive to get the include/ and lib/ directories.

Set Up Your Build System

CMake
Manual Compilation

Configure your CMakeLists.txt:

cmake_minimum_required(VERSION 3.28)
project(MyProject)

# Set paths to ONNX Runtime
set(ORT_HEADER_DIR "path/to/onnxruntime/include")
set(ORT_LIBRARY_DIR "path/to/onnxruntime/lib")

# Link against ONNX Runtime
include_directories(${ORT_HEADER_DIR})
link_directories(${ORT_LIBRARY_DIR})

add_executable(myapp main.cpp)
target_link_libraries(myapp onnxruntime)

Build your project:

cmake -S . -B build -DORT_HEADER_DIR=/path/to/include -DORT_LIBRARY_DIR=/path/to/lib
cmake --build build --config Release

Compile and link manually:

# Linux/macOS
g++ -std=c++17 main.cpp -I/path/to/onnxruntime/include \
    -L/path/to/onnxruntime/lib -lonnxruntime -o myapp

# Windows (MSVC)
cl.exe main.cpp /I"path\\to\\onnxruntime\\include" \
       /link /LIBPATH:"path\\to\\onnxruntime\\lib" onnxruntime.lib

Include the Header

In your C++ code:

#include "onnxruntime_cxx_api.h"

Make sure the ONNX Runtime shared library (.so, .dll, or .dylib) is in your system’s library path at runtime.

Build from Source

For advanced users who need custom builds:

# Clone the repository
git clone --recursive https://github.com/microsoft/onnxruntime.git
cd onnxruntime

# Build (Linux/macOS)
./build.sh --config Release --build_shared_lib --parallel

# Build (Windows)
.\\build.bat --config Release --build_shared_lib --parallel

See the build documentation for detailed instructions.

C#

ONNX Runtime is available as NuGet packages for .NET applications.

CPU
GPU (CUDA)
DirectML (Windows)
Mobile

Install the managed and native packages:

# Managed API
dotnet add package Microsoft.ML.OnnxRuntime

# Or via NuGet Package Manager
Install-Package Microsoft.ML.OnnxRuntime

The native libraries are automatically included.

For GPU support:

dotnet add package Microsoft.ML.OnnxRuntime.Gpu

For DirectML execution provider:

dotnet add package Microsoft.ML.OnnxRuntime.DirectML

For Xamarin and MAUI applications:

# iOS
dotnet add package Microsoft.ML.OnnxRuntime

# Android
dotnet add package Microsoft.ML.OnnxRuntime

The package includes targets for iOS, Android, and other mobile platforms.

Requirements

.NET 6.0 or later
Compatible with .NET Framework 4.6.2+, .NET Core 3.1+, .NET 5+
Supports Windows, Linux, macOS, iOS, and Android

Local Build

To build the NuGet managed package locally:

Windows
Linux

# Restore dependencies
msbuild -t:restore .\\src\\Microsoft.ML.OnnxRuntime\\Microsoft.ML.OnnxRuntime.csproj

# Build
msbuild -t:build .\\src\\Microsoft.ML.OnnxRuntime\\Microsoft.ML.OnnxRuntime.csproj

# Create package
msbuild .\\OnnxRuntime.CSharp.proj -t:CreatePackage -p:Configuration=Release

./build.sh --config Release --build_nuget

The .nupkg file will be in build/Release.

Java

ONNX Runtime for Java is available on Maven Central.

Maven

Add to your pom.xml:

<dependency>
    <groupId>com.microsoft.onnxruntime</groupId>
    <artifactId>onnxruntime</artifactId>
    <version>1.25.0</version>
</dependency>

For GPU support:

<dependency>
    <groupId>com.microsoft.onnxruntime</groupId>
    <artifactId>onnxruntime_gpu</artifactId>
    <version>1.25.0</version>
</dependency>

Gradle

Add to your build.gradle:

dependencies {
    implementation 'com.microsoft.onnxruntime:onnxruntime:1.25.0'
}

Requirements

Java 8 or later (Java 11+ required for building)
Compatible with Windows, Linux, and macOS

Build from Source

To build the Java binding:

# From the repository root
./build.sh --build_java --config Release

# The JAR will be in build/[OS]/Release/java/build/libs/

See the Java API build instructions for more details.

JavaScript

ONNX Runtime provides packages for Node.js and web browsers.

Node.js
Web (Browser)
React Native

Install for Node.js applications:

npm install onnxruntime-node

For development/nightly builds:

npm install onnxruntime-node@dev

The package includes pre-built binaries for:

Windows (x64, arm64)
Linux (x64, arm64)
macOS (x64, arm64)

CUDA binaries are automatically downloaded for Linux x64.

Skip CUDA Installation

To skip automatic CUDA EP installation:

npm install onnxruntime-node --onnxruntime-node-install=skip

Install for browser applications:

npm install onnxruntime-web

For WebGPU support:

npm install onnxruntime-web
# WebGPU is included in the default package

Bundle Options

ONNX Runtime Web provides multiple bundle files for different use cases:

Bundle	Size	WebGL	WASM	WebGPU
`ort.all.min.js`	682 KB	✓	✓	✓
`ort.min.js`	434 KB	✓	✓	✗
`ort.webgl.min.js`	411 KB	✓	✗	✗
`ort.webgpu.min.js`	293 KB	✗	✓	✓
`ort.wasm.min.js`	46 KB	✗	✓	✗

CDN Usage

You can also use ONNX Runtime Web from a CDN:

<script src="https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/ort.min.js"></script>

Install for React Native applications:

npm install onnxruntime-react-native

React Native requires additional setup for iOS and Android. See the React Native documentation for platform-specific configuration.

Requirements

Node.js: v20.x or later (v16+ supported)
Electron: v28.x or later (v15+ supported)
Browsers: Modern browsers with WebAssembly support

Platform-Specific Notes

Windows
Linux
macOS
Mobile

System Requirements

Windows 10 version 1809 or later
Visual Studio 2019 or later (for building from source)
Windows SDK 10.0.17763.0 or later

GPU Support

CUDA: NVIDIA GPU with CUDA 12.x
DirectML: Windows 10 version 1903 or later, DirectX 12 capable GPU

Common Issues

Install Visual C++ Redistributable if you encounter missing DLL errors

System Requirements

Ubuntu 20.04+ or equivalent
glibc 2.28 or later
GCC 9+ or Clang 10+ (for building from source)

GPU Support

CUDA: NVIDIA GPU with CUDA 12.x and cuDNN 9.x
ROCm: AMD GPU with ROCm 5.4+

Install Dependencies

# Ubuntu/Debian
sudo apt-get update
sudo apt-get install -y libgomp1

System Requirements

macOS 10.15 (Catalina) or later
Xcode 12.0+ (for building from source)

Apple Silicon

Native ARM64 builds are available for Apple Silicon (M1/M2/M3) Macs.

CoreML Support

CoreML execution provider is available on both Intel and Apple Silicon Macs:

# Python
import onnxruntime as ort
session = ort.InferenceSession("model.onnx", 
                               providers=["CoreMLExecutionProvider"])

Android

Android API Level 24 (Android 7.0) or higher
NDK r23c or later
Gradle 7.0+

Available via:

Maven: com.microsoft.onnxruntime:onnxruntime-android
AAR files from GitHub releases

iOS

iOS 12.0 or later
Xcode 14.0+
CocoaPods 1.12+

Available via:

CocoaPods: pod 'onnxruntime-c'
XCFramework from GitHub releases

Mobile builds support both ONNX format and optimized ORT format models. See ORT format documentation for details.

Verify Installation

After installation, verify ONNX Runtime is working:

import onnxruntime as ort
print(f"ONNX Runtime version: {ort.__version__}")
print(f"Available providers: {ort.get_available_providers()}")

Next Steps

Now that you have ONNX Runtime installed, check out the Quickstart Guide to learn how to run inference with your first model.

Get Started

Core Concepts

Inference

Training

Execution Providers

Performance

Model Conversion

Advanced

​Python

​Requirements

​Optional Dependencies

​C/C++

​Build from Source

​C#

​Requirements

​Local Build

​Java

​Maven

​Gradle

​Requirements

​Build from Source

​JavaScript

​Skip CUDA Installation

​Bundle Options

​CDN Usage

​Requirements

​Platform-Specific Notes

​System Requirements

​GPU Support

​Common Issues

​System Requirements

​GPU Support

​Install Dependencies

​System Requirements

​Apple Silicon

​CoreML Support

​Android

​iOS

​Verify Installation

​Next Steps

Python

Requirements

Optional Dependencies

C/C++

Build from Source

C#

Requirements

Local Build

Java

Maven

Gradle

Requirements

Build from Source

JavaScript

Skip CUDA Installation

Bundle Options

CDN Usage

Requirements

Platform-Specific Notes

System Requirements

GPU Support

Common Issues

System Requirements

GPU Support

Install Dependencies

System Requirements

Apple Silicon

CoreML Support

Android

iOS

Verify Installation

Next Steps