onnxruntime-openvino

What it is and what it does

onnxruntime-openvino is an execution provider plugin for ONNX Runtime that routes inference workloads to Intel hardware accelerators. It integrates OpenVINO inline optimizations into ONNX Runtime with minimal code changes—typically a single provider argument—enabling developers to accelerate ONNX models across Intel CPUs, integrated GPUs, discrete GPUs, and integrated NPUs without rewriting inference logic.

The package ships with prebuilt OpenVINO libraries on Linux (version 2025.4.1), eliminating separate installation steps there; Windows users must install OpenVINO separately. It supports Python 3.11–3.14 on 64-bit Linux and Windows, and depends on flatbuffers, numpy, packaging, protobuf, and sympy for runtime operation.

Use it for:

• Accelerate ONNX model inference on Intel CPUs or integrated GPUs without changing application code
• Deploy ML models to Intel discrete GPU hardware for higher throughput inference workloads
• Route inference to Intel integrated NPUs on supported platforms to reduce power consumption
• Optimize existing ONNX Runtime applications for Intel hardware with minimal refactoring

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Enables ONNX Runtime to accelerate machine learning model inference on Intel hardware (CPUs, integrated/discrete GPUs, and NPUs) using OpenVINO optimizations.

Yes, if you run ONNX models on Intel hardware and want to accelerate inference with minimal code changes. The permissive MIT license, active maintenance, and inclusion of prebuilt OpenVINO libraries on Linux reduce friction. Windows users should verify OpenVINO installation requirements before committing. No known security vulnerabilities as of the query date.

Install

onnxruntime-openvino on PyPI

pip install onnxruntime-openvino

uv add onnxruntime-openvino

poetry add onnxruntime-openvino

Installing onnxruntime-openvino

Before you install

Medium install friction with prebuilt OpenVINO libraries on Linux wheels (version 2025.4.1 included); Windows requires separate OpenVINO installation. Active maintenance with recent commits; supports Python 3.11–3.14.

License in practice

MIT License (permissive) allows commercial and private use with minimal restrictions; attribution required but no copyleft obligations.

Quickstart

pip install onnxruntime-openvino

import onnxruntime as rt
sess = rt.InferenceSession('model.onnx', providers=['OpenVINOExecutionProvider'])

Windows requires separate OpenVINO PyPI package installation; Linux wheels include OpenVINO 2025.4.1. Requires Python 3.10 or later; Ubuntu 18.04+ or Windows 10+ (64-bit).

Verify before relying

• Performance improvement magnitude across different Intel hardware types and model architectures
• Compatibility with specific ONNX opset versions or model complexity limits
• Whether discrete GPU support requires additional drivers or system configuration beyond standard Intel GPU drivers

Package facts

Evidence: onnxruntime_openvino-1.24.1-cp311-cp311-manylinux_2_28_x86_64.whl; onnxruntime_openvino-1.24.1-cp311-cp311-win_amd64.whl; onnxruntime_openvino-1.24.1-cp312-cp312-manylinux_2_28_x86_64.whl; onnxruntime_openvino-1.24.1-cp312-cp312-win_amd64.whl; onnxruntime_openvino-1.24.1-cp313-cp313-manylinux_2_28_x86_64.whl; onnxruntime_openvino-1.24.1-cp313-cp313-win_amd64.whl

Keywords: onnx, machine, learning

Tags

More Software Development packages

Further reading