This issue is tracking the status for providing a version compatible with Python 3.14. In most cases we rely on other dependencies to provide support in their distribution, where possible we will link to the corresponding issues.

Requirements for the release

• docling-core with Python 3.14 support, see feat: python 3.14 compatibility docling-core#403
• docling-parse with Python 3.14 support, see feat: support for py3.14 docling-parse#174
• docling-ibm-models with Python 3.14 support, see feat: support for python 3.14 docling-ibm-models#131
• pyarrow (needed in docling-ibm-models and vlm extra), see [Python][Packaging] Support Python 3.14 and upload wheels apache/arrow#47438
• Update lxml (see Allow lxml >= 6.0.0 #2232), feat: Support for Python 3.14 #2530
• Fix usage of torch model.compile(), see feat: Support for Python 3.14 #2530

Requirement for complete features

With this packages missing we will still make a Docling release for 3.14 but some features will not be available.

• mlx (used for vlm on mac)
• onnxruntime (used for more efficient rapidocr), see python 3.14 support microsoft/onnxruntime#26309
• numba (used in asr pipeline), see Python 3.14 numba/numba#9957
• vllm is pinning the exact torch version which is not yet 2.9 (with support for Python 3.14)
• usage of torch.compile with Python 3.14 (not yet available in torch)

Dependencies which build from source on Python 3.14

The following packaged don't provide yet compiled wheels for Python 3.14 but they can get compiled on the fly when installing them.

• scikit-image (used by easyocr), see Python 3.14 support scikit-image/scikit-image#7918
• watchdog, only for dev (it is used to build the docs pages from jupyter notebooks)
• pyclipper (used by easyocr and rapidocr)