---
id: keras-nightly
version: "3.16.0.dev2026081404"
license: Apache License 2.0
license_treatment: permissive
maintenance: active
---

# keras-nightly — Multi-backend Keras

License: permissive · Maintenance: active · Downloads: 256.6K/mo

## What it is and what it does

keras-nightly is a high-level deep learning framework designed to work with multiple compute backends—JAX, TensorFlow, PyTorch, and OpenVINO—so you can write model code once and run it on whichever backend suits your needs. It provides a unified API for building neural networks across computer vision, natural language processing, audio, time-series, and recommender systems, with the goal of letting you switch backends without rewriting your models.

The package includes 8 runtime dependencies (absl-py, numpy, rich, namex, h5py, optree, ml-dtypes, packaging) and is actively maintained with recent releases. It requires Python 3.11 or later and one or more backend packages installed separately. The framework is positioned as a drop-in replacement for tf.keras when using the TensorFlow backend, and supports consuming datasets from tf.data or PyTorch DataLoaders regardless of which backend you choose.

Use it for:

- Train a model on JAX for development speed, then switch to TensorFlow for production deployment without rewriting code.
- Build computer vision models (image classification, detection) that run on any of the four supported backends.
- Develop NLP or time-series models that can be deployed via OpenVINO for inference-only optimization.
- Convert existing tf.keras code to run on PyTorch or JAX by changing the backend environment variable.
- Write custom layers or training loops that work across multiple frameworks without framework-specific code.
- Scale model training from a laptop to GPU or TPU clusters while keeping the same high-level API.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

A multi-backend deep learning framework that lets you build and train neural networks using JAX, TensorFlow, PyTorch, or OpenVINO as the compute engine, without rewriting your model code.

Yes, if you need a unified deep learning API across multiple backends or want to avoid framework lock-in. The active maintenance, low install friction, permissive license, and zero known vulnerabilities support adoption. Install it alongside your chosen backend (TensorFlow, JAX, PyTorch, or OpenVINO) and set KERAS_BACKEND before importing. Not suitable if you require Python versions below 3.11 or need a single-backend framework.

## Install

pip install keras-nightly
uv add keras-nightly
poetry add keras-nightly

## Installing keras-nightly

Before you install: Low install friction with a pure-Python wheel. Active maintenance with a recent release and high repository engagement (64228 stars). Requires Python 3.11 or later and one or more backend packages (TensorFlow, JAX, PyTorch, or OpenVINO) installed separately.

License in practice: Apache License 2.0 (permissive) allows use, modification, and distribution with minimal restrictions, making it suitable for both open-source and commercial projects.

Quickstart:

pip install keras-nightly

import os
os.environ["KERAS_BACKEND"] = "jax"

import keras_nightly
model = keras_nightly.Sequential([keras_nightly.layers.Dense(10)])

Backend must be configured via KERAS_BACKEND environment variable before importing keras-nightly, and cannot be changed after import. A backend package (tensorflow, jax, torch, or openvino) must be installed separately.

Verify before relying:

- Performance speedup claims (20% to 350%) and their applicability to specific model architectures or use cases.
- Compatibility and testing coverage across all four backends for custom layers and training loops.
- Maturity of the OpenVINO inference-only backend relative to the other backends.

## Package facts

- License: Apache License 2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 256.6K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

multi-backend deep learning framework, neural network library, jax tensorflow pytorch unified api, high-level deep learning api, backend-agnostic model training, computer vision nlp audio models, scalable deep learning framework, multi-backend, neural-networks, framework-agnostic

[View on SkillFed](https://skillfed.io/packages/keras-nightly) · [View on PyPI](https://pypi.org/project/keras-nightly/)