Flax

Flax is Google’s open-source neural network library for JAX, designed to give researchers maximum flexibility when building and training deep learning models. It has around 7,300 stars on GitHub and remains in active development in 2026, with new commits landing this month.

Overview

JAX provides fast, composable numerical computing — automatic differentiation, JIT compilation and effortless scaling across GPUs and TPUs — but leaves the design of the neural network layer up to you. Flax fills that gap with the Linen API, a Pythonic way to define models as familiar classes while keeping JAX’s functional core: parameters are explicit, state is managed transparently, and models stay easy to transform with jit, vmap and pmap. Flax integrates with the wider Google JAX ecosystem, including Optax for optimization and Orbax for checkpointing, and it powers well-known open projects and model implementations across vision, language and reinforcement learning. Comprehensive documentation, guides and worked examples make it approachable despite its research orientation. Flax is free and open source under the Apache-2.0 license.

Key Features

• Neural network library purpose-built for the JAX ecosystem
• Linen API: intuitive model definitions with explicit, functional state handling
• Scales from a laptop to GPU/TPU pods with JAX transformations
• Integrates with Optax optimizers and Orbax checkpointing
• Used for production and research models in vision, NLP and RL
• Extensive documentation, guides and reference implementations

💳 Pricing

Flax is completely free and open source under the Apache-2.0 license. There are no paid tiers — Google maintains it as public infrastructure for the JAX community.

Who Is It For

Flax is for machine learning researchers and engineers who work in the JAX ecosystem and want a flexible, high-performance framework for training neural networks at any scale, from quick experiments to large distributed runs.