Back to Gap Map Model components / Inference code

ONNX Runtime

open source / Overall score: 4.5(leading)

ONNX Runtime is a cross-platform accelerator for machine learning inference and training, running models exported from PyTorch, TensorFlow/Keras, and classical ML libraries via the ONNX interchange format. It executes on CPUs, GPUs, NPUs, and custom accelerators through hardware-specific execution providers that target NVIDIA, AMD, Intel, Qualcomm, and Apple silicon. Microsoft develops it and uses it in production across Windows, Office, and Azure.

Openness

5 high confidence

5.0

license: MIT(OSI)
source: public(Microsoft)
core features withheld: no

Fully MIT-licensed open source from Microsoft, with no feature-gated tier.

https://github.com/microsoft/onnxruntime recorded 2026-08-09
"This project is licensed under the MIT License"; repo description "ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator"; single public repository with no separate paid or feature-gated tier referenced
https://github.com/microsoft/onnxruntime/blob/main/LICENSE recorded 2026-08-09
MIT License, Copyright (c) Microsoft Corporation, with standard MIT permission and warranty-disclaimer text
https://raw.githubusercontent.com/microsoft/onnxruntime/main/LICENSE recorded 2026-09-29
The LICENSE file reads "MIT License" followed by "Copyright (c) Microsoft Corporation" and the standard MIT permission grant.
https://raw.githubusercontent.com/microsoft/onnxruntime/main/README.md recorded 2026-09-29
The README states "This project is licensed under the [MIT License](LICENSE)." and describes the repository as a cross-platform accelerator with no separate paid or gated tier mentioned.

Adoption

5 high confidence

5.0

PyPI downloads of the onnxruntime package, over the trailing month, are ONNX Runtime's adoption signal.

https://pypistats.org/packages/onnxruntime recorded 2026-08-09
Downloads last month: 86,893,228

Capability

4 medium confidence

4.0

ONNX Runtime covers a broad set of execution providers across CPU, GPU, IoT, edge, mobile and cloud, though several are still in preview and AMD ROCm is deprecated. It is a general-purpose runtime for inference and training rather than a dedicated LLM serving engine.

https://onnxruntime.ai/docs/execution-providers/ recorded 2026-09-29
The page has a "Summary of supported Execution Providers" table listing "NVIDIA TensorRT", "Qualcomm QNN", "AMD MIGraphX" and "Android Neural Networks API" among its targets.

Verified 2026-09-24