PyTorch is an open-source machine-learning framework for building, training and running neural networks. It combines Python-friendly experimentation with tensor computation and hardware acceleration for research and production systems.
What it does Its tensor library supports CPU and accelerator workloads, while automatic differentiation computes gradients for training. Neural-network modules, optimizers and data-loading utilities provide the building blocks for custom models. Eager execution makes models easier to inspect and debug; compilation can optimize suitable workloads. Distributed tools support training across devices and machines. Python and C++ interfaces connect it to broader software stacks, with ecosystem libraries for tasks such as computer vision. It is a developer framework, not a hosted chatbot or a ready-made no-code model service.
Pricing — checked 21 October 2026 (USD) The core software is free under a BSD-style open-source license, with no per-seat subscription or platform token allowance. Running it still requires suitable hardware: local GPUs, cloud compute, storage, data transfer and commercial hosting or support are separate costs. Model weights, datasets and third-party dependencies may carry their own licenses. Foundation membership is not required to download or use the framework.
Their story Development began in 2016 with contributors from the Torch community, substantial support from Meta and participation from other organizations. PyTorch launched in 2017. In September 2022 it moved to the Linux Foundation under the PyTorch Foundation, separating broad organizational governance from technical decisions made by maintainers.
Sources: Project and license · Documentation · Foundation story