TOKYO – February 19, 2026 – Preferred Networks, Inc. (PFN) today released CuPy™ v14, a major update to the open-source library for general-purpose matrix calculation.

Having celebrated its 10th anniversary in 2025, CuPy has become a cornerstone library supporting the Python data science ecosystem. With over 30 million annual downloads and more than 10,000 GitHub stars, it continues to receive widespread support from researchers and developers worldwide.

CuPy v14 represents the first major update in two years. This version strengthens support for the latest platforms and execution environments while adding over 50 new APIs. The key features of this release are as follows:

1. Compliance with NumPy v2 Specifications

CuPy v14 aligns with the reorganized calculation rules and data type handling introduced in NumPy v2. It also implements the new set of APIs established in that version.

While it was previously possible to use CuPy and NumPy v2 in the same environment, CuPy’s internal behavior remained consistent with NumPy v1 specifications. By achieving the same semantics as NumPy v2, CuPy v14 reduces the need for users to account for differences between the two libraries, allowing for more intuitive GPU-accelerated computing.

2. Support for NVIDIA CUDA Pip Wheels

CuPy v14 now supports NVIDIA CUDA binary packages (Pip Wheels) distributed via PyPI (Python Package Index).

This eliminates the previous requirement of system-wide CUDA Toolkit installation. In addition to making CuPy easier to deploy in containerized and/or Python virtual environments, this update improves interoperability with other GPU-enabled Python libraries that support Pip Wheels, such as PyTorch.

3. Support for AMD ROCm 7

CuPy v14 now supports ROCm 7, the latest version of AMD's open-source stack for GPU computing. This ensures that users in AMD GPU environments can leverage the most up-to-date software infrastructure for their calculations.

For further details, please refer to the official CuPy blog announcement:

https://medium.com/cupy-team/announcing-cupy-v14-e8515ec05fca

PFN will continue to swiftly incorporate the latest research outcomes into CuPy while collaborating with supporting companies and open source communities.