Chainer
chainer.org
- Free plan
- Windows
- macOS
- Linux
- API
- Self-hosted
7.4 Club score Visit chainer.org
What the maker publishes
- Automatic differentiation
- It provides automatic differentiation APIs using the define-by-run approach, also known as dynamic computational graphs. source
- Community support
- The documentation links to Chainer community Slack chat and forums. source
- Development status
- Chainer is in maintenance phase, with further development limited to bug fixes and maintenance. source
- Extensions
- ChainerRL implements deep reinforcement-learning algorithms, and ChainerCV provides tools for computer-vision neural networks. source
- Founded
- 2015 source
- GPU computing
- Chainer supports CUDA computation and can run on multiple GPUs; CUDA support requires installing CuPy separately. source
- GPU support
- Chainer supports CUDA computation and multi-GPU use; CUDA support requires CuPy, and cuDNN support is optional. source
- Installation
- The documentation recommends installing Chainer with pip and also describes installation from a source tarball or Git repository. source
- Integrations
- The v7 release announcement says ONNX-Chainer is integrated into Chainer and that most Chainer features, including ChainerMN, are compatible with ChainerX ndarray. source
- License
- The GitHub repository identifies Chainer's license as the MIT License. source
- Maintenance status
- Chainer is in maintenance, with further development limited to bug fixes and maintenance. source
- Migration
- The maker says it chose PyTorch as Chainer's replacement and released migration resources for users moving to PyTorch. source
- Model types
- It supports feed-forward, convolutional, recurrent, recursive, and per-batch network architectures. source
- Network architectures
- It supports feed-forward, convolutional, recurrent, and recursive networks, as well as per-batch architectures. source
- Optional capabilities
- Optional dependencies enable image-dataset support, HDF5 serialization, and distributed deep learning with ChainerMN. source
- Optional features
- Image dataset support, HDF5 serialization, and distributed deep learning with ChainerMN require optional dependencies. source
- Platform limits
- The documentation recommends 64-bit Ubuntu 14.04 or 16.04 and CentOS 7, and says it cannot guarantee operation on Windows or macOS. source
- Platforms
- The recommended environments are 64-bit Ubuntu 14.04/16.04 LTS and CentOS 7; the documentation says it cannot guarantee operation on Windows or macOS. source
- Product
- Chainer is a Python-based, standalone open-source framework for deep learning models. source
- Product
- Chainer is a Python-based, standalone open-source framework for building deep-learning models. source
- Python control flow
- Forward computation can use Python control-flow statements while retaining backpropagation, which the site says makes code intuitive and easy to debug. source
- Python support
- Chainer v7 supports Python 3.5.2 and later versions through 3.8, while Python 2 is unsupported. source
- Target users
- The maker describes Chainer as supporting deep-learning research and development, and says it chose PyTorch as the replacement framework for its own development efforts. source
Checked against the maker's pages on 11 Oct 2026.