2026.4

Bodo 2026.4 Release (Date: 04/09/2026)

🎉 Highlights

This release, we are excited to introduce Bodo’s [CPU–GPU hybrid execution model][df_gpu] available via conda (currently only supports Linux/Python 3.14):

conda install -c bodo.ai -c rapidsai -c nvidia "bodo=*=*cuda" --no-channel-priority

With the CUDA-enabled variant of the Bodo package, supported DataFrames operations will automatically run on all available GPUs, while unsupported operations seamlessly fall back to CPU.

The CUDA-enabled Bodo package uses CUDA-aware OpenMPI for GPU communication. Before running CPU–GPU hybrid code, you must first set OpenMPI’s communication layer to UCX:

export OMPI_MCA_pml=ucx

✨ CPU–GPU hybrid execution

The following types of operations will run on GPUs:

• Parquet read (local filesystem, S3, Google Cloud Storage, HDFS, Azure Data Lake)

• Parquet write (local filesystem, S3, Google Cloud Storage, HDFS, Azure Data Lake)

• Most kinds of joins

• Vectorized filters and projections (except UDFs)

• Column arithmetic and boolean ops (except UDFs)

• GroupBy aggregations: sum, count, mean, min, max, var, std, size, skew, nunique

• Series reductions: sum, product, count, mean, min, max

• drop_duplicates, concat, Series.isin

🏎️ Performance Improvements

• Improved join ordering.

🛠️ Infrastructure

• Added Docker ARM build.

⚙️ Dependency Changes

• Upgraded DuckDB planner.

Full Changelog: 2026.2...2026.4

2026.2

What's Changed

• build platform conda package for Python 3.13 by @scott-routledge2 in #1002
• Update Cluster Configs for TPCH SF1000 by @scott-routledge2 in #1008
• Support Pandas 3rc1 by @ehsantn in #1003
• gpu join by @DrTodd13 in #985
• Build with OpenMPI on Mac by @scott-routledge2 in #1010
• Remove Vendored MPI4py from pip builds by @scott-routledge2 in #1011
• Add remaining Pandas on Spark Queries to TPCH benchmark by @scott-routledge2 in #1012
• Smarter way to determine CPU or GPU usage. by @DrTodd13 in #1013
• fix wheel build problem by @DrTodd13 in #1016
• Infrastructure for async GPU pipeline. by @DrTodd13 in #1015
• BSE-5277: GPU shuffle by @IsaacWarren in #1017
• Async gpu write parquet by @DrTodd13 in #1018
• [BSE-5279] Multi Rank CPU-GPU support by @scott-routledge2 in #1014
• Fix GPU Join Min/Max stats on multiple ranks by @IsaacWarren in #1020
• Aync gpu project and gpu_filter. by @DrTodd13 in #1019
• GPU filter read parquet by @DrTodd13 in #1021
• Start transfer to GPU earlier via GPUBatchGenerator by @scott-routledge2 in #1023
• Update TPCH Readme with distributed results by @scott-routledge2 in #1026
• Write parquet recreate directory. by @DrTodd13 in #1025
• Speed up build by @IsaacWarren in #1027
• Configure and build separately by @IsaacWarren in #1028
• Support Pandas 3 by @ehsantn in #1009
• Multi-GPU Join by @IsaacWarren in #1024
• Add rapids include by @IsaacWarren in #1032
• Upgrade Pyiceberg Dependency by @scott-routledge2 in #1031
• Upgrade to PyArrow 23 by @scott-routledge2 in #1033

Full Changelog: 2026.1...2026.2

2026.1

What's Changed

• 2025.12.2 Release Notes by @scott-routledge2 in #992
• Use consistent variables in TPCH by @scott-routledge2 in #994
• BSE-5236: Join Filter IO Followup by @IsaacWarren in #997
• Fix cross compile by @IsaacWarren in #999
• Linux Arm CI and Packaging by @scott-routledge2 in #993
• Fix typo in oracle install on Nightly by @scott-routledge2 in #1000
• Support Pandas 3rc1 (first PR) by @ehsantn in #996
• BSE-5250: Windows 3.14 Pip Wheels by @IsaacWarren in #1001

Full Changelog: 2025.12.2...2026.1

2025.12.2

What's Changed

• Add release notes for 2025.12.1 by @IsaacWarren in #990
• Estimate Parquet Dataset Cardinality for Plan by @scott-routledge2 in #989
• Upgrade to Python 3.14 by @ehsantn in #979
• Exclude Windows Python 3.14 Pip wheel by @scott-routledge2 in #991
• BSE-5236: Join Filter IO Pushdown by @IsaacWarren in #987

Full Changelog: 2025.12.1...2025.12.2

2025.12.1

What's Changed

• Handle tuple column selection in df.loc by @ehsantn in #968
• Fix Conda Build Pipeline by @scott-routledge2 in #966
• Skip tests to unblock PR CI by @scott-routledge2 in #967
• Avoid double cast for integer division by @ehsantn in #970
• Support negative n in df.head() by @ehsantn in #972
• 2025.12 Release Notes by @scott-routledge2 in #971
• Fix slicing with negative step by @ehsantn in #973
• Keep input timestamp unit in Series.dt.tz_localize by @ehsantn in #974
• Fix testing issues on Nightly by @scott-routledge2 in #969
• Cast binary op result to output type by @ehsantn in #976
• Fix output Series.name for pd.to_datetime() by @ehsantn in #977
• Support concurrent async messages to the same rank in shuffle by @scott-routledge2 in #975
• Skip Iceberg DDL tests on PR CI by @scott-routledge2 in #980
• TPCH benchmarking improvements by @scott-routledge2 in #964
• Avoid using arrow types for to_datetime by @scott-routledge2 in #983
• Upgrade to Numba 0.63.1 by @ehsantn in #978
• AWS titan-text eol by @IsaacWarren in #984
• Fixes for TPCH scripts by @scott-routledge2 in #986
• Fix overflow in parquet read cardinality est by @scott-routledge2 in #988

Full Changelog: 2025.12...2025.12.1

2025.12

Bodo 2025.12 Release (Date: 12/11/2025) {#December_2025}

🎉 Highlights

This release, we are excited to add Join Filters in the plan optimizer, significantly improving performance on real workloads. We also improve Bodo's timezone support and fix several minor bugs.

✨ New Features

• Support datetime.datetime in query plans.
• Validate repl argument of Series.str.replace same as Pandas.
• Improve concat output order.
• Support Series.take().
• Support timezones in to_datetime.

🏎️ Performance Improvements

• Add Join Filters to remove rows with keys that won’t match a join key as early as possible.
• Box/unbox date arrays using Arrow.
• Box/unbox time arrays using Arrow.

🐛 Bug Fixes

• Avoid hang in scattering BodoSeries.
• Fix explode bug for Arrow large list type.
• Allow int64/uint64 mismatch in some internal data structures.
• Fix for ArrowExtensionArray iloc indexing.
• Fix timezone in convert_dtypes.
• Fix for nested join.
• Fixed remove unused column pass to keep column references alive in conjunction with duckdb version upgrade.
• Better support for operations that result in empty dataframes.

Full Changelog: 2025.11.2...2025.12

2025.11.2

What's Changed

• Pipeline reordering. by @DrTodd13 in #933
• NaN/NA handling. by @DrTodd13 in #926
• 2025.11 Release notes by @IsaacWarren in #936
• Numba 0.62.1 upgrade by @scott-routledge2 in #934
• Add metrics for Iceberg Read/Parquet Read by @scott-routledge2 in #935
• More fixes for narwhals tests. by @DrTodd13 in #937
• Filter manifest files based on partition summaries by @scott-routledge2 in #938
• Pin Numba to 0.62 by @scott-routledge2 in #940
• Increase timeout for BodoSQL smoke test by @IsaacWarren in #939
• Support np.ufunc calls on BodoSeries by @ehsantn in #941
• Todd/prune fix by @DrTodd13 in #942

Full Changelog: 2025.11.1...2025.11.2

2025.11.1

What's Changed

• Support groupby.size() (with no value cols) by @scott-routledge2 in #904
• Don't redistribute data twice in example by @IsaacWarren in #912
• Fix test_basic_iceberg_read path issue on nightly by @ehsantn in #916
• CTE column pruning by @DrTodd13 in #910
• First attempt at adding Narwhals to test suite. by @DrTodd13 in #918
• Avoid JIT imports in BodoSQL C++ backend by @ehsantn in #907
• Support passing BodoDataFrames to BodoSQL C++ backend without extra plan executions by @ehsantn in #919
• Timezone Support by @scott-routledge2 in #915
• Support creating empty DataFrames by @ehsantn in #921
• Fixes for test_unique narwhals test by @scott-routledge2 in #920
• Change param name to not conflict on windows. by @DrTodd13 in #924
• Expose pandas.Timestamp in bodo.pandas by @scott-routledge2 in #922
• Fix BodoSQLContext use inside JIT by @ehsantn in #928
• Fix drop_duplicates() for non-trivial Indexes by @ehsantn in #925
• Run Narwhals tests on a single worker and fix issues by @scott-routledge2 in #923
• BSE-5174: Duckdb Planner Upgrade by @IsaacWarren in #917
• Support specifying Glue Catalog in pd.read_sql_table by @scott-routledge2 in #929
• BSE-5206: Fix windows duckdb by @IsaacWarren in #930
• Support passing BodoSQL as an arg to JIT by @scott-routledge2 in #932

Full Changelog: 2025.11.0...2025.11.1

2025.11.0

What's Changed

• Add release notes for 2025.10.1 by @scott-routledge2 in #890
• Fix docker release files symlink target by @IsaacWarren in #894
• TPCH improvements. by @DrTodd13 in #893
• Support TPC-H Q5 in BodoSQL C++ backend by @ehsantn in #886
• Support filesystem Iceberg catalog in BodoSQL C++ backend by @ehsantn in #895
• Support Iceberg filter/project/limit in BodoSQL C++ backend by @ehsantn in #898
• Skip pandas ddp example by @IsaacWarren in #892
• Improvements based on Narwhals tests. by @DrTodd13 in #897
• Capture API usage. by @DrTodd13 in #896
• Add all duckdb timestamp types by @IsaacWarren in #900
• Call tokenize with just a tokenizer by @IsaacWarren in #901
• Support join filters in BodoSQL C++ backend by @ehsantn in #902
• Finetune on Iceberg Data by @IsaacWarren in #903
• Update demo notebook by @scott-routledge2 in #908
• Fix two narwhals issues. by @DrTodd13 in #906
• Upgrade to arrow 22 by @scott-routledge2 in #905
• Remove Python 3.9 in a few places by @ehsantn in #911
• Add iceberg marker to test by @ehsantn in #913

Full Changelog: 2025.10.2...2025.11.0

2025.10.2

What's Changed

• Remove reindex check in publish_binary scripts by @scott-routledge2 in #889
• Refactor Guides Docs by @scott-routledge2 in #882
• BSE-5132: prepare_dataset by @IsaacWarren in #873

Full Changelog: 2025.10.1...2025.10.2