Performance History
This page tracks the performance of different solvers over time. This can be used to see which solvers are improving, and on what kinds of benchmarks. Once again, you can filter the benchmark set to your problems of interest and the graphs will automatically re-generate to show you the performance history on your chosen subset.
Our platform includes the highest version of each solver released on conda each year since 2020. For 2025, we include a solver version if the solver had a major or minor release in 2025 as of April 20, 2025. The 2025 results will be updated with the last version released in 2025 at the end of this year. Some solver versions are not available on conda or have compatibility issues with our benchmarking infrastructure, see here for more details. (GLPK has not had a release since 2020.)
Relative Solver Performance Evolution
The below graphs show the shifted geometric mean (SGM) of the runtime / memory consumption each solver version on all the benchmarks in the (filtered) set of instances that all solver versions have been run on. For each year, the SGMs have been normalized so that the best performing solver in that year has a value of 1.0. Therefore, if a solver version has (normalized) SGM runtime of 1.25, it indicates that it is on average 25% slower than the best performing solver in that specific year.
Caveats
Some solvers returned errors when running on some benchmark instances. For more details, please see the tracking issue.