AI Potluck
Contribute
Back to Gap Map Infrastructure / Classic ML & computer vision

AutoGluon

AutoGluon
open source / Overall score: 3.8

Automated machine-learning library that trains, tunes and stack-ensembles models for tabular prediction, time-series forecasting and multimodal image, text and tabular tasks from a few lines of code. Its tabular predictor fits gradient-boosted trees, neural networks and tabular foundation models such as Mitra, TabICL and TabPFN side by side, and its forecaster includes Chronos. Developed at Amazon, which also offers it as a managed service.

Openness

5 high confidence
5.0
license
Apache-2.0(OSI)
source
public(github.com/autogluon/autogluon)
core features withheld
no — AutoGluon Cloud and SageMaker Autopilot are separate hosted offerings

Apache-2.0, and every module - tabular, time series and multimodal - builds from the public repository. Amazon's hosted ways to run it, AutoGluon Cloud and SageMaker Autopilot, are offered as separate services around the same open library. Some of the third-party foundation models it can call carry their own non-commercial terms.

Adoption

3 medium confidence
3.0

Measured on monthly PyPI downloads of autogluon.core, the package every AutoGluon module depends on, so that installs of the tabular, time-series and multimodal parts are each counted once; the autogluon meta-package alone would undercount.

Capability

5 high confidence
5.0

AutoGluon does the whole prediction task for you: it chooses and trains candidate models, including pretrained tabular foundation models, and stacks them into an ensemble, across tables, time series and mixed image and text data. It is the most complete automation in this category and the point the other products here are compared against.

Verified 2026-09-26