Quantum optimization research now has a shared benchmark for testing progress against strong classical methods. The Quantum Optimization Benchmarking Library, or QOBLIB, covers ten difficult problem classes with common metrics, baselines, and public result submissions.
IBM described the latest release in an August 12 update, alongside publication in Nature Computational Science and a new public website. More than 2,000 submitted results already appear in the library.
Ten problems, one comparison framework
Optimization papers often choose different problem instances, hardware settings, and classical baselines. A favorable setup sometimes produces an impressive result without answering whether a quantum method beats the best classical alternative.
QOBLIB addresses this comparison problem. Its ten problem classes form what the project calls an intractable decathlon. Each class becomes difficult for leading classical solvers at relatively small sizes while staying within reach of near-term quantum hardware.
The library supplies:
- Standard problem definitions and instance generators
- Metrics for comparing solution quality and runtime
- Classical and quantum baselines
- Tools for submitting and reproducing results
- Public records for positive and negative findings
Researchers from Zuse Institute Berlin, Purdue University, the National University of Singapore, IBM, E.ON, Kipu Quantum, Forschungszentrum Jülich, and other organizations contributed to the project.
Why classical baselines matter
A quantum optimization result needs more than a successful circuit. Researchers must compare the result against modern classical solvers using equal problem data and meaningful resource accounting.
Our QAOA guide shows how quantum approximate optimization works on Max-Cut. Small demonstrations help explain the algorithm, yet classical solvers still dominate practical problem sizes. QOBLIB gives researchers a common place to track where this gap narrows or grows.
Negative results also belong in the library. A failed quantum run prevents other teams from repeating weak approaches and improves the record of current hardware limits.
QOBLIB does not prove advantage
The benchmark organizes evidence. QOBLIB itself does not demonstrate quantum advantage in optimization. IBM presents the library as infrastructure for finding and verifying such a result.
Problem selection still matters. Hardware access, noise, runtime accounting, and classical solver tuning all influence comparisons. Public methods and shared instances make those choices easier to inspect.
What changes now
Quantum optimization claims gain a clearer test. New algorithms enter the same problem classes and face published baselines. Classical improvements enter the record too, preventing old comparisons from setting an easy target.
QOBLIB started as a preprint and open-source repository in 2025. The 2026 Nature publication, public website, and result volume move the project from proposal toward community infrastructure.