NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #4809 most downloaded on PyPI
Access Kaggle resources anywhere
Latest stable
2.2.5
Last release 1 months ago
22 Oct 2026
Release timing varies
gaps range from 9 days to 7 months
Most releases are documented
notes for 40 of the last 60 stable releases
2 versions withdrawn
withdrawn after publishing
9 years old
93 releases · first in 2018
Drag across the ruler to compare, or tap a release to open it.
Finish 2.2.5 release prep and bump twine to 7.0.0
Finish 2.2.5 release prep and bump twine to 7.0.0 (#1230)
Follow-up to #1228 before triggering the 2.2.5 release:
- Bump `__version__` in `src/kaggle/__init__.py` to `2.2.5`
- Sync `kagglesdk==0.1.38` in `requirements.lock`
- Restore `### Next` heading in `CHANGELOG.md` and update `kagglesdk`
version note
- Bump `twine` to `7.0.0`, drop `pkginfo`, and regenerate
`tools/releases/requirements.txt`
certifi >= 2026.10.19, which drops the compromised emSign/eMudhra root CAs (GHSA-6q48-6c7w-9gw2). kaggle only pulled in certifi through requests, which sets no minimum, so an existing install could keep a certifi that would accept a forged kaggle.com certificate and leak the user's API token to a man-in-the-middlekaggle kernels delete instead of failing with a ValueError when a version is included in <owner>/<kernel>/<version>kaggle competitions submit with a competition-specific hint on stderr, naming the notebook and version for code submissions. The check never matched a 404, so users got the generic <owner>/<slug> hint insteadkaggle competitions submit when the submission file upload fails. It printed "Could not submit to competition" and exited 0, so scripts treated the failed submission as a success+00:00) in refreshAt timestamps displayed by kaggle quotapipx or uv and add troubleshooting guidance for externally-managed-environment errors in the installation docskaggle kernels output files to disk through the same download path as datasets and competitions, with a read timeout, a size check, and resume after a dropped connection. Each file used to be read into memory whole, so a 512 MiB output took about 1.2 GB of RAM, and a dropped connection failed the command and left an empty file behind<owner>/<kernel>/<version> for kaggle kernels pull, output, status, files and logs, which previously returned 403 for pull and acted on the latest version for the others--accelerator help and kernels docs that NvidiaTeslaT4 maps to GPU T4 x2 and TpuV5E8 maps to TPU v5e-8kaggle datasets delete to prevent unintentionally deleting an entire dataset when targeting a single versionkaggle kernels delete, including whitespace-only ones, to prevent unintentionally deleting an entire kernel when targeting a single version, and stop advertising a version in the kernels delete and kernels topics list helpkaggle competitions submission <ref>, which already comes back in the API response but was not displayed-k/--kernel in kaggle kernels pull, which was ignored when the positional kernel argument was omitted, causing the CLI to fall back to the local kernel-metadata.json and pull a different kernelkaggle kernels output file whose server-supplied name resolves outside the requested --path directory--no-run to kaggle kernels push to save a new version without executing the notebook, the equivalent of Quick Save in the web UI--unzip when downloading a single file with kaggle datasets download -f, which extracts the file from the zip archive the server wraps large files in, and add --unzip to kaggle competitions downloadkaggle competitions submit when --kernel, --version, or --file is missing for a code competition submissionkaggle datasets download -f or kaggle competitions download -f, instead of writing it to the download rootslug instead of title in the kaggle datasets create error message when the requested slug is already in useRange downloads from overwriting existing partial download fileskaggle kernels output files so error responses are not saved as output fileskaggle competitions host-add <comp> -u <user> to grant host access on a competition to a Kaggle user--debug flag), and list common examples in kaggle --helpkaggle models instances versions download --untar against path traversal0700) and verify directory ownership and permissions before usekaggle benchmarks quota to show Model Proxy (AI inference) spend quota, and bump kagglesdk to >= 0.1.38--untar in kaggle models instances versions download when the archive is already cached locallykaggle kernels output to avoid UnicodeEncodeError on Windowsoverview and usage fields from the update mask when updating a model instancekaggle competitions submission-download <id> to download the submitted file for a single submission (requires kagglesdk >= 0.1.36)deadline (Competition Deadline) to the competition settings command and bump kagglesdk to >= 0.1.36kaggle kernels push requests a retired accelerator (NvidiaTeslaP100, TpuV38, Tpu1VmV38, TpuV232 or TpuV2256), since the server runs the session on a replacement instead. Drop the retired shapes from the docsLLMS_AVAILABLE token reference in the benchmarks documentationkaggle search command across competitions, datasets, notebooks, models, users, and discussionss, m, h, d, w) in kaggle auth print-access-token --expiration instead of raising a TypeError--wait/--poll-interval to kaggle competitions submit to wait for scoring, and add kaggle competitions submission <ref> to look up a single submission's status and scoreN/A instead of Fail in kaggle benchmarks leaderboard for a task result the server returned without a pass/fail or numeric score, which was reported as a failed taskYour coding agent can read these notes before it upgrades. Set up the MCP server →
fix(benchmarks): support owner/task separator in benchmark commands
Update --format help text to reference output_format.md
kaggle kernels logs --follow to use SSE log stream (#999)Clarify LLMS_AVAILABLE vs. full model set in benchmarks docs
As titled. For kaggle-cli launch
Release 2.2.1 (#1044)
As titled. For kaggle-cli launch
kaggle competitions team-submissions command (#1036)kaggle quota command for GPU/TPU accelerator quota (#1029)fix(benchmarks): use repeated -m/-d flags instead of space separated …
fix(benchmarks): use repeated -m/-d flags instead of space separated …
…values (#1028)
follow up on
#1019 (comment)
* Update kagglesdk version
fix(benchmarks cli): bugs and additional features
kaggle b init to include example and reference (#990)b auth and b init confirmations (#986)--format flag to datasets status for JSON output (#972)Add kaggle benchmarks init command
kaggle benchmarks init command (#981)Add kaggle benchmarks auth command
kaggle benchmarks auth command (#976)kaggle kernels logs CLI command (#966)dataset-cover-image.png upload for datasets metadata --update (#969)Release 2.0.1, see CHANGELOG.md for details.
Release 2.0.1, see CHANGELOG.md for details.
--sandbox flag to kaggle competitions submit for sandbox submissions (competition hosts/admins only) (#932)datasets metadata --update:
Change more "instance" to "variation"
Rename kaggle-api to kaggle-cli
kaggle-api to kaggle-cliAdd version checking against server known-version
* Changes to build script
Add machine_shape to the metadata of kaggle kernels pull
Nothing published for this version
Fix a problem in downloading kernel output files.
Fix a dataset download problem. Datasets that had a license were failing to download.
Version 1.7.3 was never released. There were errors in versioning on test.pypi.org. For consistency, we decided to jump several version numbers. This
Version 1.7.3 was never released. There were errors in versioning on test.pypi.org. For consistency, we decided to jump several version numbers. This is the first release since 1.6.17.
The actual changes are described in 1.7.3.
Fix build issues after changing driver from setup to pip
Fix build issues after changing driver from setup to pip (#703)
The next beta release has already been published.
I had some strange errors, but I wasn't first. Upgrading `pkginfo` and
`twine` fixed things.
However, projects that use kaggle/api/kaggle_api.py may be affected. That file is deprecated and will be removed. Most of its functions still work, bu…
kaggle/api/kaggle_api.py may be affected. That file is deprecated and will be removed.
Most of its functions still work, but those that involve uploading files no longer work.
The command-line tool uses a higher-level abstraction for uploading and client code needs
to be converted to use that.No changes; release 1.6.16 did not complete.
Support XDG base directory specification on Linux
No changes; release 1.6.13 isn't usable.
Re-release 1.6.11 without the src directory included in the package.
src directory included in the package.Allow unauthenticated usage of "datasets download", "datasets files".
Repackage of 1.6.8 as a new release, to fix the problematic 1.6.9 release.
Repackage of 1.6.8 as a new release, to fix the problematic 1.6.9 release.
Do not use. Problematic release that causes an error: ModuleNotFoundError: No module named 'kaggle.api'
ModuleNotFoundError: No module named 'kaggle.api'* Add "gguf"
Add "GemmaCpp" and "GGML" model frameworks.
* Add "MaxText" model framework.
Add "Transformers" model framework.
Add "Flax" and "Pax" frameworks
Release date: 01/11/24
Release date: 01/09/24
Release date: 01/11/24
Release date: 01/09/24 * Add "Other" model framework.
Release date: 01/09/24
Release the pre-release branch with models endpoints.
Release date: 01/08/24
Release date: 01/04/24
Release the pre-release branch with models endpoints.
Release date: 01/04/24
Release date: 11/22/23
Release date: 9/19/23
Release date: 8/02/23
Release date: 7/17/23
Release date: 7/07/23
-y to delete model/instance/version without confirmationRelease date: 7/06/23
Release date: 6/30/23
-dRelease date: 6/29/23
Release date: 6/12/23
Release date: 6/07/23
Release date: 2/27/23
Nothing published for this version
Nothing published for this version
Nothing published for this version
Alpha Release for Models endpoints
Release date: 7/06/23
Nothing published for this version
Nothing published for this version
Fix dataset download bug with locale
Release date: 7/17/23
Add missing licenses for datasets
Release date: 6/30/23
-dShow the full error message from the API
Release date: 6/29/23
Nothing published for this version
Release date: 03/12/21 * No changes
Release date: 03/12/21
Release date: 11/30/20 * Remove dependency on slugify.
Release date: 11/30/20
Drop version restriction on urllib3 in setup.py.
Release date: 10/21/20
…names have been changed to snake_case. This is a breaking change for anyone directly using the python code rather than simply using the command line.
Release date: 09/03/20
Release date: 8/31/20
Release date: 9/19/19
Release date: 8/30/19
Release date: 5/28/19
Release date: 2/20/19
Release date: 1/28/19
Release date: 12/5/18
Release date: 12/5/18
Release date: 10/19/18
Release date: 8/28/18
Release date: 8/28/18
-p argument actually optinalresources field when updating a datasetcode_file value in kernel metadata when pulling kernelsRelease date: 8/7/18
Release date: 8/1/18
Release date: 7/30/18
Release date: 7/30/18
-c, -d, and -k argumentsresources field in dataset metadataRelease date: 7/20/18
Release date: 7/20/18
Release date: 7/19/18
Release date: 6/25/18
'KAGGLE_CONFIG_DIR' environmental tokenRelease date: 6/12/18
Release date: 6/10/18
Release date: 6/9/18
Release date: 6/8/18
Release date: 5/18/18
Release date: 5/18/18
kaggle c is the same thing as kaggle competitions)Release date: 5/7/18
Release date: 5/4/18
Release date: 4/30/18
columns to fieldsRelease date: 4/26/18
Release date: 4/24/18
Release date: 4/24/18
Release date: 4/19/18
Release date: 4/18/18
Release date: 4/17/18
camelCasing other than class names have been changed to snake_case. This is a breaking change for anyone directly using the python code rather than simply using the command line.Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per quarter.