Netron is a free, open-source visualizer for neural network and machine learning models with 33,500+ stars on GitHub — drag in a model file and see its graph.
Model files are opaque binaries until you can see their structure. Netron renders them as interactive graphs: layers, operators, weights shapes and metadata, all explorable in the browser at netron.app or in desktop apps for macOS, Windows and Linux. It understands an unusually wide range of formats — ONNX, TensorFlow and TensorFlow Lite, Keras, PyTorch, Core ML, scikit-learn, safetensors and many more — which makes it the universal inspector when converting models between frameworks or debugging why an export went wrong. Created by Lutz Roeder and maintained for years, it is one of those quiet infrastructure tools nearly every ML engineer has used. The project is MIT licensed and updated in October 2026.
Its zero-install web version is a quiet superpower for collaboration: a link to netron.app lets a colleague inspect the exact model you are debugging without setting up an environment, and the exported graphs double as documentation in papers, blog posts and design reviews. Teachers use it to make abstract architectures concrete for students, and MLOps teams use it as a final visual check in conversion pipelines moving models from training frameworks to edge runtimes like TensorFlow Lite and Core ML. Few tools are simultaneously this simple and this universal.
Netron is completely free and open source under the MIT license, including the hosted viewer at netron.app.
ML engineers, researchers and students who need to inspect, debug, teach or present neural network architectures across any framework.