How Good are Learned Cost Models, Really? Insights from Query Optimization Tasks

Summary: Systematic evaluation of Learned Cost Models for query optimization across main tasks: join ordering, access path, and operator choice. Seven LCMs vs traditional models; surprisingly, traditional cost models often beat LCMs, guiding future work. (summarized by gpt-5-nano on Feb 09 2026)

Paper ID: hf7d7ee140c7e4899
Venue: SIGMOD
Year: 2025
Pagerank: 6.0392183e-05
Overall Rank: 5,683 | 61.80%
DOI: 10.1145/3725309

Incoming Non-self Citations Over Time

Authors

1. Roman Heinrich (German National Research Center for Information Technology; Technical University of Darmstadt)
2. Manisha Luthra (German National Research Center for Information Technology; Technical University of Darmstadt)
3. Johannes Wehrstein (Technical University of Darmstadt)
4. Harald Kornmayer (Baden-Wuerttemberg Cooperative State University)
5. Carsten Binnig (German National Research Center for Information Technology; Technical University of Darmstadt)

BibTeX Citation

@inproceedings{heinrich_sigmod25,
        title = {{How Good are Learned Cost Models, Really? Insights from Query Optimization Tasks}},
        author = {Heinrich, Roman and Luthra, Manisha and Wehrstein, Johannes and Kornmayer, Harald and Binnig, Carsten},
        series = {{SIGMOD} '25},
        booktitle = {Proceedings of the {ACM} {SIGMOD} International Conference on Management of Data},
        publisher = {Association for Computing Machinery},
        doi = {10.1145/3725309},
        url = {https://dl.acm.org/doi/10.1145/3725309},
        year = {2025}
}

Incoming Citations (Sorted by Pagerank)

Showing 11 of 11 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank
10,290	This is Going to Sound Crazy, But What If We Used Large Language Models to Boost Automatic Database Tuning Algorithms By Leveraging Prior History? We Will Find Better Configurations More Quickly Than Retraining From Scratch!	2026	SIGMOD	5.0431863e-05
10,350	Survivorship Bias in Industrial Database Workloads	2026	CIDR	4.9793485e-05
10,412	Are Learned DBMS Components Robust to Workload Drift?: [Experiments & Analysis]	2026	SIGMOD	4.9793485e-05
10,679	Reqo: A Comprehensive Learning-Based Cost Model for Robust and Explainable Query Optimization	2026	SIGMOD	4.9793485e-05
10,700	Sample-based Distinct Cardinality Estimation for Multiple Attributes in Multi-Dataset Queries	2026	VLDB	4.9793485e-05
10,741	OBELISK: Efficient Offline Query Planning with Bayesian Optimization-Informed Language Model Reasoning	2026	VLDB	4.9793485e-05
10,955	How Out-of-Bounds Are Your Cardinality Estimates?	2026	VLDB	4.9793485e-05
11,015	Vodka: Rethink Benchmarking Philosophy in HTAP Systems	2026	VLDB	4.9793485e-05
11,055	SafeLoad: Efficient Admission Control Framework for Identifying Memory-Overloading Queries in Cloud Data Warehouses	2026	VLDB	4.9793485e-05
11,388	Opening The Black-Box: Explaining Learned Cost Models For Databases	2025	VLDB	4.9793485e-05
11,425	Learned Cost Models for Query Optimization: From Batch to Streaming Systems	2025	VLDB	4.9793485e-05

Outgoing Citations (Sorted by Pagerank)

Showing 28 of 28 cited papers.

Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.

Rank	Cited Paper	Year	Venue	Pagerank
1	Access Path Selection in a Relational Database Management System	1979	SIGMOD	0.0023947656
15	How Good Are Query Optimizers, Really?	2016	VLDB	0.00061066921
85	Learned Cardinalities: Estimating Correlated Joins with Deep Learning	2019	CIDR	0.00035864347
318	DeepDB: Learn from Data, not from Queries!	2020	VLDB	0.00021167555
362	Bao: Making Learned Query Optimization Practical	2021	SIGMOD	0.00019989474
406	Deep Unsupervised Cardinality Estimation	2020	VLDB	0.00019045544
461	An End-to-End Learning-based Cost Estimator	2020	VLDB	0.00017829982
512	NeuroCard: One Cardinality Estimator for All Tables	2021	VLDB	0.00017050173
560	Plan-Structured Deep Neural Network Models for Query Performance Prediction	2019	VLDB	0.00016403151
629	Performance Prediction for Concurrent Database Workloads	2011	SIGMOD	0.00015429949
1,064	Are We Ready For Learned Cardinality Estimation?	2021	VLDB	0.00012202282
1,199	Balsa: Learning a Query Optimizer Without Expert Demonstrations	2022	SIGMOD	0.00011563985
1,839	Why TPC Is Not Enough: An Analysis of the Amazon Redshift Fleet	2024	VLDB	9.5304799e-05
2,210	Lero: A Learning-to-Rank Query Optimizer	2023	VLDB	8.8257742e-05
2,250	QueryFormer: A Tree Transformer Model for Query Plan Representation	2022	VLDB	8.7533306e-05
2,342	Learned Cardinality Estimation: An In-depth Study	2022	SIGMOD	8.6060437e-05
2,522	Learned Cardinality Estimation: A Design Space Exploration and A Comparative Evaluation	2022	VLDB	8.3477168e-05
2,842	Query Performance Prediction for Concurrent Queries using Graph Embedding	2020	VLDB	7.949193e-05
2,885	Zero-Shot Cost Models for Out-of-the-box Learned Cost Prediction	2022	VLDB	7.9094988e-05
3,316	Cloud Analytics Benchmark	2023	VLDB	7.4373161e-05
4,202	Kepler: Robust Learning for Faster Parametric Query Optimization	2023	SIGMOD	6.7374091e-05
4,258	LEON: A New Framework for ML-Aided Query Optimization	2023	VLDB	6.6994722e-05
5,481	A Comparative Study and Component Analysis of Query Plan Representation Techniques in ML4DB Studies	2024	VLDB	6.1125124e-05
5,649	Sample-Efficient Cardinality Estimation Using Geometric Deep Learning	2024	VLDB	6.052326e-05
5,840	Pre-training Summarization Models of Structured Datasets for Cardinality Estimation	2022	VLDB	5.9737703e-05
6,308	Is Your Learned Query Optimizer Behaving As You Expect? A Machine Learning Perspective	2024	VLDB	5.8177833e-05
6,416	A Unified Transferable Model for ML-Enhanced DBMS	2022	CIDR	5.7920805e-05
7,033	Rethinking Learned Cost Models: Why Start from Scratch?	2023	SIGMOD	5.6168499e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	1,064	Are We Ready For Learned Cardinality Estimation?	2021	VLDB
2	461	An End-to-End Learning-based Cost Estimator	2020	VLDB
3	3,210	Efficiently Approximating Selectivity Functions using Low Overhead Regression Models	2020	VLDB
4	9,333	Deep Query Optimization	2019	SIGMOD
5	6,308	Is Your Learned Query Optimizer Behaving As You Expect? A Machine Learning Perspective	2024	VLDB
6	7,033	Rethinking Learned Cost Models: Why Start from Scratch?	2023	SIGMOD
7	4,741	Machine Learning for Databases	2021	VLDB
8	2,885	Zero-Shot Cost Models for Out-of-the-box Learned Cost Prediction	2022	VLDB
9	2,690	Cost-based or Learning-based? A Hybrid Query Optimizer for Query Plan Selection	2022	VLDB
10	11,425	Learned Cost Models for Query Optimization: From Batch to Streaming Systems	2025	VLDB