ONNX provides an open source format for AI models, covering deep learning and traditional machine learning. It defines an extensible computation graph model, built-in operators, standard data types, and a common file format so AI developers can use models across frameworks, tools, runtimes, and compilers, with current focus on inferencing.