ONNX Runtime

license

MIT(OSI)

source

public(Microsoft)

core-gated

ungated

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

• https://github.com/microsoft/onnxruntime/blob/main/LICENSE recorded 2026-06-04

  MIT License text, Microsoft Corporation

~74.7M PyPI downloads in the last month for the onnxruntime package; GitHub reports 'used by' ~89.9k repositories. Embedded across Microsoft products and the broader ML deployment stack. Well into the >10M band on download volume. (Note: onnxruntime spans general ML inference, not only LLM serving; see capability/recategorize note.)

• https://pypistats.org/packages/onnxruntime recorded 2026-06-04

  ~74.7M downloads last month

• https://github.com/microsoft/onnxruntime recorded 2026-06-04

  20.7k stars; 'used by 89.9k'; cross-platform inference accelerator description

Strong, MLPerf-present general-purpose inference accelerator, but its MLPerf coverage is vision/classical models, not LLM-serving throughput; for transformer/LLM serving it trails the purpose-built vLLM/SGLang/TensorRT-LLM (C5) frontier. Scored 4 within this category. Arguably broader than inference_code (a general ML runtime), but it does load weights and serve predictions, so it fits the litmus test; flagging for reviewer consideration without forcing a move.

• https://github.com/mlcommons/mlperf_inference_test_submissions_v5.0 recorded 2026-06-04

  ONNX Runtime (v1.20.1-v1.23.2) submissions for ResNet50/RetinaNet offline

• https://github.com/microsoft/onnxruntime recorded 2026-06-04

  cross-platform inference accelerator, plugin execution providers (CUDA/WebGPU)