Benchmarks¶

This page lists benchmark results obtained by recent versions of the PyVRP package and reference VRP solvers.

Hint

See the benchmarking instructions for details on how we benchmark.

PyVRP¶

The table below contains the benchmark results obtained by each new significant version of PyVRP. The reported values represent the percentage gap to the best-known solutions at the time of the benchmark, averaged over ten seeds. These best-known solutions are tracked in the PyVRP/Instances repository.

Note

We always benchmark against the most recent best-known solutions.

Date	Version	CVRP	VRPTW	PCVRPTW	MDVRPTW	VRPB
22 March 2024	0.8.0	0.23%	0.58%	0.39%	0.99% [1]	0.35%
29 January 2024	0.7.0	0.23%	0.48%	0.21% [2]	0.44%
31 August 2023	0.6.0	0.24%	0.54%
1 August 2023	0.5.0	0.22%	0.40%
9 July 2023	0.4.2	0.18%	0.43%
20 May 2023	0.3.0	0.25%	0.43%
13 May 2023	0.2.0	0.25%	0.43%
28 February 2023	0.1.0	0.22%	0.45%

Reference VRP solvers¶

The table below contains the benchmark results obtained by reference VRP solvers. The reported values represent the average gaps to the best-known solutions (at the time of the benchmark) over ten seeds.

Date	Name	CVRP	VRPTW	PCVRPTW	MDVRPTW	VRPB
11 February 2024	Google OR-Tools [3]	5.23%	10.86%	13.24%
28 February 2023	HGS-CVRP	0.11%
28 February 2023	HGS-DIMACS		0.32%
2020	ILS-SP					1.09% [4]
2013	HGS-ADC				0.71% [5]

Notes