Overview

AutoKeras is a Python package for automating parts of machine-learning model design and selection. It is aimed at people comfortable building workflows in code, especially those working with image, text, or structured data. Its appeal is configurable model search at no software cost, rather than a point-and-click workflow.

Key features

The package supports classification and regression for images, text, and structured data. That range makes it a practical option for teams handling several common supervised-learning tasks within Python. Its documentation also includes examples for multi-task and multimodal datasets, although using those capabilities remains part of a code-based workflow.

AutoKeras lets users assemble custom models from its blocks while searching for detailed configurations. For image work, architecture type, normalization, and augmentation can be chosen directly or left to the search process. This offers more control than handing the entire design over to automation, but requires users to understand the model choices they are making.

The best model found can be exported as a Keras Model. By default, AutoKeras reserves the final 20% of training data for validation and selects models by validation loss. Users can set another split, provide validation data themselves, and customize the metric or objective used for selection. When overwrite is false, rerunning the same code continues the previous fit from the directory contents, which is useful for resuming work rather than starting again.

Installation is through the Python package with pip instructions provided by the project. The installation guide supports Python 3 and recommends PyTorch 2.3.0 or later when using the PyTorch backend. Questions are directed to GitHub Discussions. DATA Lab at Texas A&M University develops the package with keras-team for version 1.0 and above.

Pricing

AutoKeras is free and open source. The AutoKeras plan costs 0.00 USD per free and consists of the Python package installed with pip. There is no paid tier described, so the trade-off is not a restricted entry plan versus an upgrade: users get the package without a software charge, while taking responsibility for running a code-based, self-hosted workflow.

Platforms

AutoKeras is available for Linux, macOS, and Windows, with self-hosted deployment. Its workflow interface is code, so it suits users who want to work in Python rather than manage models through a web interface. Feature engineering and automated model selection are supported.

Who it's for

AutoKeras is a strong fit for Python practitioners who want automated configuration search but still need to shape models from reusable blocks or control image-processing choices. It is less suitable for readers seeking a hosted visual interface or a workflow that avoids package installation and coding.

Pros and cons

Pros: It covers image, text, and structured-data classification and regression in one package, reducing the need to switch tools across these task types.
Pros: Users can combine custom blocks with automated search, then export the selected model as a Keras Model.
Pros: The free, open-source package supports resuming a fit and changing validation and model-selection settings.
Cons: The code-only, self-hosted approach puts setup and workflow management on the user; it is not a web-based option.
Cons: The default validation split uses the last 20% of training data, so users who need a different split must configure one or provide validation data.
Cons: The installation guide recommends PyTorch 2.3.0 or later for the PyTorch backend, a consideration for users planning that setup.

Alternatives

AutoML Software is a broader category directory for comparing tools across the field.

Consider FEDOT for another free, open-source AutoML framework available on Linux, macOS, Windows, and self-hosted setups, as well as through an API.

LightAutoML is another free Python library installable from PyPI, with Linux, macOS, Windows, web, and self-hosted platform options.

MLJAR AutoML offers a free open-source Python package, and also a separate Free plan with 50 prompts per month and 10 published co…; consider it when those plan options and its web availability fit your workflow.

AutoGluon is another free, open-source Python library for Linux, macOS, Windows, and self-hosted use.

Choose BigML if you prefer a web or API service: its free monthly plan includes unlimited tasks and storage, but caps datasets at 16 MB per task, allows two parallel tasks, and is limited to one user.

Amazon SageMaker Autopilot is a paid API and web option with pay-as-you-go pricing and no minimum fees or upfront commitments.

Auto-PyTorch is a free, self-hosted option for Linux, developed by the AutoML Groups of the University of Freiburg and Hannover.

JADBio suits readers looking for a web-accessible or API-based option; its free Basic plan is limited to one seat, three projects, 50 MB uploads, 500 MB storage, and one model export.

Verdict

Choose AutoKeras if you work in Python and want no-cost model search with room to define parts of the architecture and export the result as a Keras Model. Its main strength is that balance of automation and control across several data types. Look elsewhere if you need a visual hosted workflow or want to avoid managing a self-hosted package.

	AutoKeras	Obviously AI
Free plan	Yes	Yes
Free trial	No	No
Paid from	Free	Free
Open source	No	No
Platforms	Linux, macOS, self-hosted, Windows	Web
Feature engineering	Yes	—
Automated model selection	Yes	—
Workflow interface	code	—
Hosting model	self_hosted	—
Free plan	—	Yes

AutoKeras vs Obviously AI

AutoKeras

Obviously AI