Keras

What is Keras?

Keras is a high-level deep learning API for building, training and shipping models. Keras 3 supports multiple backends, including JAX, TensorFlow, PyTorch and MLX, so a model can move across frameworks.

Key capabilities of Keras

• Multi-backend API: one Keras codebase that runs on JAX, TensorFlow, PyTorch or MLX.
• Model building APIs: Functional API and subclassing for defining models.
• Built-in training: compile() and fit() for training workflows.
• KerasHub: Keras 3 implementations of popular architectures with pretrained checkpoints.
• Pretrained models: featured families include Gemma, Qwen, GPT-OSS and Mistral.
• Production export: move trained models toward deployment.

How Keras works

You define a model with the Functional API or by subclassing, choose a backend, then train with compile() and fit(). KerasHub provides ready-made architectures with pretrained checkpoints on Kaggle Models, which you load with a preset and use for training or inference on the supported backends.

Who uses Keras?

Keras is used by machine learning engineers, researchers and students who want a readable API on top of their preferred framework. Teams that want to avoid lock-in to one backend benefit from the multi-backend design.

Keras pricing

Keras is an open-source library, and the project site lists no paid plans. Costs come from the compute you train and serve on.

Keras alternatives

Modular MAX is an inference platform for serving models, so it covers deployment rather than model building. Nebius and Vast.ai rent GPU infrastructure for training and inference. DeepInfra and fal host models behind APIs, which is a different path from writing and training models yourself.