August 6th 2026 by Frank Hutter, Marius Lindauer, Katharina Eggensperger and Matthias Feurer
After a decade, we are closing automl.org.
This is not the end of a research field; it is the end of a website that outlived the constellation of groups it was built to represent. When the domain first went live in 2014, "AutoML" was a word that needed explaining. It named a workshop, a handful of tools, and the conviction of a small group in Freiburg that the tedious, expert-dependent parts of machine learning could and should be automated. Today, AutoML has its own conference, its own textbooks, its own industrial products, and its own generation of researchers. A single lab website is no longer the right container for any of that, and with automl.space, we found a new, open home.
The three groups behind automl.org have also moved on. Frank's research now also runs through Prior Labs and the ELLIS Institute Tübingen; Katharina has moved from a research group leader in Tübingen to building a new chair at TU Dortmund and the Lamarr Institute; Marius' group in Hannover has grown into an institute with a far broader objective than AutoML alone. That field didn't stay in one place and moves with the people who built it, thus, keeping a shared "Freiburg - Hannover - Tübingen" front page would no longer be accurate. We would rather close it honestly, say what automl.org achieved, and point to where the work continues.
It began as infrastructure for a workshop. The AutoML workshop at ICML ran annually from 2014, with the site hosting each edition, and by 2019 the workshop had grown to more than a hundred participants, one of the largest at ICML. Around that scaffolding, the site accumulated everything else a young field needed, and nobody else was providing: tool documentation, literature overviews, benchmark descriptions, best-practice guides, a blog that explained new papers in plain language, and a book.
The first ChaLearn AutoML Challenge (2015–2016) was where the field's claims met a scoreboard. Team AAD Freiburg entered with a new system built around scikit-learn, Bayesian optimization via SMAC, meta-learning-based warmstarting, and post-hoc ensembling. It placed in the top three in nine of the ten phases and won six of them, taking both the automatic and the human-tweakable tracks in the final two rounds. That system was auto-sklearn, and it went on to become one of the most widely used open-source AutoML packages in the world at that point in time.
The second ChaLearn AutoML Challenge (2017–2018) confirmed it was not a fluke. PoSH-Auto-sklearn, i.e., portfolio successive halving, built by Matthias Feurer, Katharina Eggensperger, Stefan Falkner, Marius Lindauer and Frank Hutter, finished ahead of all 41 other participating systems.
Beyond AutoML proper, methods and tools developed in the groups contributed to prize-winning entries in international competitions on SAT solving and AI planning, and the labs continued to compete in later challenges, including a podium finish in the NeurIPS 2022 AutoML Decathlon. But the two ChaLearn victories remain the ones that mattered most: they turned an argument into evidence.
If the competitions made the case, the code made it usable. SMAC and its successor SMAC3 for algorithm configuration and Bayesian optimization; BOHB and HpBandSter, then DEHB and NePS, for multi-fidelity optimization; and the flagship end-to-end AutoML System auto-sklearn, as well as extensions and follow-ups, including Auto-PyTorch and auto-sklearn2.0. Furthermore, fANOVA, HPOBench, NASLib, DACBench, CARL and DeepCAVE, as well as several other libraries and tools for the essential work of defining configuration spaces, efficient benchmarking, analysis and interpretability. Most of it still lives at github.com/automl, and it is not going anywhere.
The most recent chapter of that lineage is TabPFN, the prior-data fitted network for tabular data that was published in ICLR in 2023 and Nature in January 2025 and has since been downloaded millions of times.
Automated Machine Learning: Methods, Systems, Challenges (Hutter, Kotthoff and Vanschoren, Springer 2019) was the field's first book, and we made it freely available from day one. The AutoML MOOC followed, along with tutorials at ICML, NeurIPS, IJCAI, AAAI and ECML, and an annual AutoML School that has now run six times, virtually in 2021, then Freiburg, Munich, Hannover, Tübingen, and this year (2026) Dortmund.
In 2022, after eight years, we retired the workshop series and launched the International Conference on Automated Machine Learning. The first edition drew 170 people. The fifth takes place in Ljubljana this autumn, organized by a committee that reaches far beyond our supergroup, which is exactly what we hoped would happen.
The most durable output of a research group is never a paper or a package. It is people. A partial roll call of those who passed through:
In academia. Frank Hutter is co-founder and CEO of Prior Labs, spun out of the Freiburg lab in late 2024, and acquired by SAP in 2026 to anchor a European frontier lab for structured data, as well as Hector Endowed Fellow and PI at the ELLIS Institute Tübingen and Professor at the University of Freiburg. He won an ERC starting grant, an ERC POC grant and an ERC Consolidator grant.
Marius Lindauer joined Franks group in Freiburg in 2014 and became a tenure-track professor in 2019 at the Leibniz University Hannover and after winning an ERC Starting Grant was promoted to Full Professor, later got an ERC POC and was founding director of LUH|AI, the institute of AI in Hannover. This first branching off from Freiburg was the moment a single group transformed into a supergroup spanning more than one university.
Katharina Eggensperger, who did her PhD in Frank's group and during that time built efficient surrogate benchmarks, co-developed auto-sklearn and contributed to TabPFNv1, moved on to build a research group on AutoML for Science at the University of Tübingen from 2023; she is now Professor for Machine Learning and Artificial Intelligence at TU Dortmund University and the Lamarr Institute for Machine Learning and Artificial Intelligence.
Matthias Feurer, also a graduate of Frank's group and the main author of auto-sklearn, is now an assistant professor (German: Juniorprofessor) for Automated Machine Learning and Optimization at TU Dortmund University and the Lamarr Institute, after being a Thomas Bayes Fellow of the Munich Center for Machine Learning and interim professor at LMU Munich.
Many other members also moved on to build their own academic research teams.
Aaron Klein, who also did his PhD in Frank's group, built the AutoML junior research group at ScaDS.AI in Leipzig after years as a senior scientist on AWS SageMaker, and now works at the ELLIS Institute Tübingen, co-leading the OpenEuroLLM push there with David Salinas. Jan N. van Rijn, a former PostDoc, is now an Associate Professor at Leiden University and co-leads the ADA group there. André Biedenkapp will soon start his own Emmy-Noether research group at the KIT in Karlsruhe. Theresa Eimer leads a research group at the L3S Research Center and the reinforcement learning team in Hannover. Marcel Wever has just been awarded one of five AI research groups in Lower Saxony to work on energy-efficient foundation models. Josif Grabocka, who led the Representation Learning lab alongside us in Freiburg, is Full Professor of Machine Learning at the University of Technology Nuremberg.
In industry. Many graduates and former members moved on to successful careers in industry. Thomas Elsken at Tensordyne after being at Bosch, Stefan Falkner is at Bosch Corporate Research, Ilya Loshchilov at NVIDIA, Sam Müller transitioned to DataDog, after being at Meta, Noah Hollmann co-founded Prior Labs and serves as its CTO, Rhea Sukthanker is a Senior Researcher at Microsoft and Alexander Tornede is at Beckhoff Automation. Others are at Amazon, Google DeepMind, Microsoft, Meta, or further startups.
And many more than we can list here: dozens of PhD students, postdocs, research engineers and master's students whose names are on the papers, the commits and the competition entries. Thank you. You were the point.
To everyone who read the blog, filed an issue, cited a benchmark, came to a school, sent a student, or argued with us at a poster: Thank you. It was a privilege to host the front door for a while.
Frank, Marius, Katharina and Matthias