Cleanlab
Cleanlabcleanlab is a Python library that finds problems in machine-learning datasets using the outputs of a model trained on them. Its Datalab interface flags label errors, outliers, near duplicates and other issues in text, image, audio and tabular data, and the package also trains noise-robust classifiers and scores annotator quality when several people labeled the same data. It implements confident learning, a published method for estimating label noise.
Openness
5 high confidence- license
- Apache-2.0 (OSI)
- source
- public (cleanlab/cleanlab)
- core features withheld
- no — the library is the whole product; Cleanlab's commercial offering is a separate platform for checking AI agent responses, and the acquisition letter commits the package to stay open source
cleanlab is Apache-2.0 and the whole library is in the public repository. The company's paid product is a separate platform for catching mistakes in AI agent responses, not an edition of this package, so nothing in the library is held back.
- https://cleanlab.ai/ recorded 2026-11-12
The company site sells a platform to "Detect and remediate incorrect responses from any AI Agent", deployed by VPC or SaaS, and carries the banner "Cleanlab has been acquired by Handshake AI."
- https://cleanlab.ai/blog/handshake-acquires-cleanlab/ recorded 2026-11-12
A letter dated January 28, 2026 announces that Handshake acquired Cleanlab and tells users "Cleanlab's data-centric AI package will remain open-source under a more permissive Apache-2.0 license."
- https://pypi.org/pypi/cleanlab/json recorded 2026-11-12
PyPI lists cleanlab 2.9.0, uploaded 2026-01-13, with the license field "Apache License Version 2.0, January 2004".
- https://raw.githubusercontent.com/cleanlab/cleanlab/master/LICENSE recorded 2026-11-12
The root LICENSE is the stock Apache License 2.0 text, ending with the unfilled appendix line "Copyright [yyyy] [name of copyright owner]".
- https://raw.githubusercontent.com/cleanlab/cleanlab/master/README.md recorded 2026-11-12
The README presents "Cleanlab's open-source library" and a "Run cleanlab open-source" section installing from PyPI or source, and says "All features of cleanlab work with any dataset and any model". Searched the body for enterprise, studio, paid and license key: none.
Adoption
2 medium confidenceAdoption is the trailing month of PyPI downloads of cleanlab. Conda installs, which the documentation also offers, are not counted.
- https://pypistats.org/api/packages/cleanlab/recent recorded 2026-11-12
last_month 51,918, last_week 10,911, last_day 3,061 for the cleanlab package.
Capability
3 medium confidencecleanlab finds mislabeled examples, outliers and duplicates across most supervised tasks using confident learning, a peer-reviewed method. It audits labels rather than producing them, a step below Snorkel Flow, which also builds labels programmatically and runs annotation and evaluation workflows.
- https://arxiv.org/abs/1911.00068 recorded 2026-11-12
The confident learning paper (JAIR) reports "CL performance exceeding seven recent competitive approaches for learning with noisy labels on the CIFAR dataset" and says "These results are replicable using the open-source cleanlab release."
- https://docs.cleanlab.ai/stable/index.html recorded 2026-11-12
The documentation says the package checks a dataset for "label errors, outliers, near duplicates, IID violations, underperforming groups" with Datalab, working "with any ML model by operating on model outputs".
- https://raw.githubusercontent.com/cleanlab/cleanlab/master/README.md recorded 2026-11-12
The README lists dedicated support for binary and multi-class, multi-label and token classification, regression, image segmentation, object detection, multi-annotator data, active learning and outlier detection.
Verified 2026-11-12