3.4.0 (Aug 4, 2026)

XGBoost 3.4.0 completes the vector-leaf implementation for the hist tree method, revises the quantile regression and mean absolute error (MAE) objectives, removes column-split support, and fixes issues across multiple components.

Vector Leaf and Multi-output Models

The vector-leaf implementation for the hist tree method is now feature-complete and classified as experimental rather than work in progress. Key changes include:

• Categorical features on CPU and GPU, DART, monotonic and interaction constraints, and min_child_weight. (#12305, #12299, #12276, #12340, #12341, #12294, #12325, #12296)

• Distributed training and Dask multi-label estimators, including AUC and SHAP support, plus tree dumping and DataFrame conversion. (#12292, #12369, #12313, #12314, #12293)

• Batched split application, reduced-gradient sampling, shared histogram building and split evaluation, additional validation, and external-memory fixes. (#12365, #12321, #12397, #12387, #12336, #12330, #12320, #12312)

Quantile Regression and Absolute Error

Quantile regression and mean absolute error now use smooth approximations instead of line searches to determine leaf values. (#12391, #12386, #12373, #12346)

Breaking Changes and Deprecations

• Column-split support has been removed. (#12333, #12363, #12354)

• The federated learning plugin is no longer included in Python binary wheels. (#12376)

• The random-forest wrappers are deprecated. Use num_parallel_tree directly for random-forest models. (#12324, #12342)

• The deprecated XGDMatrixCreateFromFile function has been removed. (#12297)

• The CUDA asynchronous memory pool is now recommended for external-memory training. (#12337)

Build and Platform

• The default xgboost binaries for the Python and JVM packages are now built with CUDA 13.3 on Linux and Windows. Python users who need CUDA 12.9 can install xgboost-cu12 from PyPI or build from source. (#12384, #12394)

• Both the xgboost and xgboost-cpu packages support Windows on AArch64. (#12394)

Optimizations

• Reduce allocations in the JSON text parser. (#12317)

• Skip per-tree depth computation for single-row prediction. (#12307)

• Avoid rebuilding the regular expression when parsing the device parameter. (#12306)

• Extend the depth bucket used by GPU Quadrature TreeSHAP. (#12290)

Bug Fixes

The following fixes affect all interfaces. Interface-specific fixes are listed in the corresponding package sections.

• Fix an issue where min_child_weight could produce an empty root node. Also, this release fixes its interaction with max_delta_step. (#12322, #12296)

• Preserve tiny positive Hessians during GPU quantization. (#12266)

• Fix compatibility with RMM 26.08. (#12316, #12269)

• Fix learning-to-rank pair sampling when continuing training. (#12332)

• Fix GPU SHAP on CUDA SM120. (#12368)

• Fix the recoder’s handling of UTF-8 data and pandas nullable categorical indexes, and add overflow checks. (#12371)

Python Package

• Migrate Python package builds to scikit-build-core and include license files in both installed packages and wheel metadata. (#12219, #12349, #12280)

• trees_to_dataframe() now uses pandas NA for unavailable vector-leaf values instead of a mix of np.nan and None. (#12293)

• Correct scikit-learn input tags and raise ValueError when qid is omitted from score(). (#12383, #12335)

• Pass Graphviz keyword arguments through to_graphviz(). (#12359)

• Fix shape reporting for empty array-interface results and SciPy CSC input handling. (#12381, #12378)

• Fix allreduce() always returning a flattened one-dimensional array. (#12377)

• Code cleanup. (#12331)

R Package

• Fix early_stopping_rounds handling in xgboost(). (#12370)

• Remove an obsolete 32-bit Windows thread_local workaround. (#12352)

• Test and lint fixes. (#12309, #12259)

JVM Packages

• Fix potential resource leaks in native error paths and make floating-point serialization locale-independent. (#12286, #12235)

• Update the Maven Central publishing plugin and release tooling, including support for local testing. (#12255, #12254, #12253)

Documentation and Demos

• Add the second-place solution from the Tabular Playground Series. (#12392)

• Fix typos and update external links. (#12400, #12401, #12357, #12358, #12328)

CI and Maintenance

• Update GitHub Actions. (#12375, #12338, #12300, #12263, #12274)

• Add a stale pull request cleanup workflow. (#12318)

• Clean up the NCCL discovery script. (#12262)

• Refactor and clean up code. (#12345, #12275, #12385, #12271)

• Release maintenance. (#12252)