High-level, multi-backend deep-learning API for building, training and deploying neural-network models.
Keras is an open-source Python library that abstracts the low-level details of modern deep-learning frameworks and lets you create, train and deploy neural-network models with just a few lines of code.
Since Keras 3, the same model code can run on top of TensorFlow, JAX or PyTorch, giving developers the freedom to pick the backend that offers the best performance or ecosystem fit. The API provides a large suite of modular building blocks—layers, losses, optimizers, metrics and callbacks—plus batteries-included features such as mixed-precision training and multi-device distribution.
Beyond the core API, the Keras ecosystem includes Keras Applications (pre-trained computer-vision networks), Keras Tuner (hyper-parameter search) and Keras Hub (a growing catalog of state-of-the-art pretrained checkpoints). These tools, combined with Keras’ focus on readable, production-ready code, have made it a go-to choice for researchers and companies alike—from NASA and CERN to self-driving-car teams and large-scale recommendation systems.
Whether you are prototyping on a laptop or training at scale on GPUs, TPUs or CPUs, Keras delivers the same concise developer experience and seamless model portability—all while remaining 100 % open source and backed by an active community led by its creator, François Chollet.
