Updates
Any person or group can submit improved solutions. After the solutions are checked, they will be included in CVRPLib. On the other hand, claims of new optimal solutions will only be accepted if a citable reference for the method used is provided. Each change in the status of an instance will be recorded chronologically on this page.
November 3, 2026
by CVRPLib
Forty-six improved solutions for the VRPTW reported by Mario (DataFirst Logistics): XTW-n280 (17 / 43,314), XTW-n289 (7 / 24,571), XTW-n308 (9 / 27,191), XTW-n331 (24 / 52,473), XTW-n336 (30 / 71,673), XTW-n351 (22 / 26,665), XTW-n376 (5 / 26,567), XTW-n384 (16 / 34,990), XTW-n401 (35 / 79,439), XTW-n420 (41 / 44,641), XTW-n429 (15 / 34,602), XTW-n439 (18 / 32,857), XTW-n449 (24 / 55,684), XTW-n469 (13 / 40,142), XTW-n480 (7 / 28,180), XTW-n491 (22 / 33,134), XTW-n502 (46 / 87,501), XTW-n524 (24 / 44,470), XTW-n536 (58 / 64,413), XTW-n561 (23 / 32,816), XTW-n573 (3 / 23,926), XTW-n586 (30 / 56,364), XTW-n599 (76 / 105,076), XTW-n613 (24 / 33,485), XTW-n627 (17 / 41,499), XTW-n641 (10 / 36,653), XTW-n655 (31 / 36,695), XTW-n670 (11 / 36,942), XTW-n685 (71 / 76,503), XTW-n701 (26 / 60,459), XTW-n716 (59 / 85,912), XTW-n733 (25 / 39,532), XTW-n749 (26 / 39,150), XTW-n766 (58 / 123,694), XTW-n783 (6 / 43,505), XTW-n801 (50 / 93,666), XTW-n819 (46 / 63,175), XTW-n837 (49 / 84,586), XTW-n856 (48 / 64,369), XTW-n876 (44 / 85,127), XTW-n895 (22 / 50,626), XTW-n916 (19 / 48,206), XTW-n936 (19 / 42,454), XTW-n957 (75 / 89,137), XTW-n979 (89 / 209,914) and XTW-n1001 (39 / 80,544). DataFirst Logistics, own solver.
Three improved solutions reported by Chris Pawlowski (dayjob.ai): XL-n4063-k347 (548,287), XL-n4738-k487 (759,603) and XL-n6588-k473 (333,806) using SISR and AILS system.
November 2, 2026
by CVRPLib
One improved solution for the VRPTW reported by Jean-Michel Landry (UQAM): R1_10_6 (46,860.3) using hgs-advance (Vidal 2026) for extended run with no early stop.
One improved solution reported by F.P.Saccomanno (University of Calabria): X-n957-k87 (85,464) using iterated local search.
November 2, 2026
by Rafael Martinelli
The detailed rules of the VRPTW BKS Challenge are out! The Challenge will run on XLTW, a new VRPTW set with 1,000 to 10,000 customers built from the XL instances of the 2026 CVRP Challenge. In the meantime, the XLTW-Test set, with similar instances, is available for algorithm development and testing. Rules, scoring and timeline are here.
And a surprise: the XTW set is now part of CVRPLib! These 100 VRPTW instances extend the CVRP X set with time windows. XTW is not part of the Challenge, but its instances already come with feasible solutions, 22 of them proven optimal, and you can submit improved solutions right now.
Also new on the site:
- Interactive plot viewer: route table with show/hide and highlight, customer and depot tooltips, arrows and export. For the VRPTW it adds route schedules, a timeline for each route and a time slider that replays the solution.
- Archives for every set: one with the instances and one with the BKS of every instance, always up to date.
- BKS history: each instance has a page with its bound history, and the plot page shows the BKS details (optimal, date, authors).
- Instance conventions: instances that differ from their set’s conventions (explicit distance matrix, maximum route duration, service time) are listed on the problem, instances and plot pages.
November 1, 2026
by CVRPLib
Nine improved solutions reported by Arhan Canli (Quantavane): Leuven2 (111,353), Antwerp1 (477,256), Antwerp2 (291,252), Ghent1 (469,481), Ghent2 (257,631), Brussels1 (501,427), Brussels2 (344,811), Flanders1 (7,238,613) and Flanders2 (4,364,450) using HGS-CVRP (Vidal) to re-optimise groups of nearby routes, starting from the published BKS.
One improved solution for the VRPTW reported by Arhan Canli (Quantavane): C1_10_9 (40,278.7) using PyVRP (Wouda, Lan and Kool) to re-optimise groups of nearby routes, starting from the published BKS.
Four improved solutions reported by Chris Pawlowski (dayjob.ai): XL-n2121-k186 (283,092), XL-n7516-k859 (572,939), XL-n8575-k1297 (1,088,340) and XL-n8766-k1032 (905,019) using SISR and AILS system.
Three improved solutions for the VRPTW reported by Jean-Michel Landry (Université du Québec à Montréal): R1_10_1 (52,982.0), R1_10_5 (50,365.5) and R1_10_9 (49,088.1) using HGS (Vidal) variant.
Two improved solutions reported by Chris Pawlowski (dayjob.ai): XL-n5406-k783 (1,040,099) and XL-n8960-k634 (772,866) using SISRS and AILS system.
October 31, 2026
by CVRPLib
Two improved solutions reported by Laio Oriel Seman (IFSC): XL-n5288-k1246 (1,957,938) and XL-n8389-k2028 (3,352,773). using a modified version of HGS: the incumbent is seeded with the current BKS and improved by repeatedly re-solving small neighborhoods of routes with independent, parallel HGS populations under threshold acceptance and diverse elite restarts.
Four improved solutions reported by Chris Pawlowski (dayjob.ai): XL-n3147-k232 (256,399), XL-n5288-k1246 (1,957,929), XL-n5774-k290 (249,846) and XL-n8389-k2028 (3,352,681) using SISR and AILS system.
October 30, 2026
by CVRPLib
Two improved solutions reported by Laio Oriel Seman (IFSC): XL-n2961-k55 (107,934) and XL-n10001-k1570 (2,329,417). using a modified version of HGS: the incumbent is seeded with the current BKS and improved by repeatedly re-solving small neighborhoods of routes with independent, parallel HGS populations under threshold acceptance and diverse elite restarts.
October 29, 2026
by CVRPLib
Two improved solutions reported by Laio Oriel Seman (IFSC): XL-n5288-k1246 (1,957,944) and XL-n10001-k1570 (2,329,438), using a modified version of HGS: the incumbent is seeded with the current BKS and improved by repeatedly re-solving small neighborhoods of routes with independent, parallel HGS populations under threshold acceptance and diverse elite restarts.
One improved solution reported by Christoph Breunig (LYKOPHOS): XL-n2961-k55 (107,936). Starting from the published BKS, subsets of routes are re-solved with HGS-CVRP (Vidal 2022) using threshold acceptance and an elite pool.
Two improved solutions reported by Chris Pawlowski (dayjob.ai): XL-n8389-k2028 (3,352,774) and XL-n10001-k1570 (2,329,443) using AILS and SISR system.
October 28, 2026
by CVRPLib
Two improved solutions reported by Laio Oriel Seman (IFSC): XL-n5288-k1246 (1,957,947) and XL-n10001-k1570 (2,329,471), using a modified version of HGS: the incumbent is seeded with the current BKS and improved by repeatedly re-solving small neighborhoods of routes with independent, parallel HGS populations under threshold acceptance and diverse elite restarts.
October 28, 2026
by Rafael Martinelli
GREAT NEWS: We've just wrapped up the beta of CVRPLib's new automatic BKS publishing pipeline! Researchers submit improving solutions directly through the site, they're verified on the spot, and a daily digest publishes everything automatically. No more waiting on manual review!
Check and publish single solution: https://galgos.inf.puc-rio.br/cvrplib/en/check/single
Check and publish an archive of solutions: https://galgos.inf.puc-rio.br/cvrplib/en/check/archive
October 27, 2026
by CVRPLib
Five improved solutions reported by Arhan Canli: XL-n1048-k237 (380,090), XL-n1748-k271 (173,756), XL-n5774-k290 (249,847), XL-n8028-k294 (266,253) and XL-n10001-k1570 (2,329,784).
October 27, 2026
by CVRPLib
Five improved solutions reported by Arhan Canli: XL-n1048-k237 (380,090), XL-n1748-k271 (173,756), XL-n5774-k290 (249,847), XL-n8028-k294 (266,253) and XL-n10001-k1570 (2,329,784).
Two improved solutions reported by Chris Pawlowski (Dayjob.ai): XL-n10001-k1570 (2,329,483) and XL-n10001-k1570 (2,329,546) using SISR and AILS system.
Two improved solutions reported by Chris Pawlowski (Dayjob.ai): XL-n10001-k1570 (2,329,483) and XL-n10001-k1570 (2,329,546) using SISR and AILS system.
October 26, 2026
by CVRPLib
One improved solution reported by Christoph Breunig (LYKOPHOS): XL-n2121-k186 (283,097) using as start the published BKS, subsets of routes are repeatedly re-solved with HGS-CVRP, using threshold acceptance and an elite pool.
October 25, 2026
by CVRPLib
Two improved solutions reported by Christoph Breunig (LYKOPHOS): X-n895-k37 (53,845) and X-n979-k58 (118,923) using as start the published BKS, subsets of routes are repeatedly re-solved with HGS-CVRP, using threshold acceptance and an elite pool.
Three improved solutions reported by Arhan Canli: XL-n6168-k1922 (1,527,161), XL-n9784-k2774 (4,076,853) and XL-n10001-k1570 (2,329,785).
Three improved solutions reported by Arhan Canli: XL-n6168-k1922 (1,527,161), XL-n9784-k2774 (4,076,853) and XL-n10001-k1570 (2,329,785).
October 24, 2026
by CVRPLib
Two improved solutions reported by Martín Vizzolini (Independent Researcher): X-n1001-k43 (72,332) and XL-n10001-k1570 (2,329,795), starting with the published BKSs and improved on top of them.
October 23, 2026
by CVRPLib
Thirteen improved solutions for the VRPTW reported by Seiya Takahashi, Tsubasa Kumamoto, Ryotaro Shin, Masahiro Nomura, Isao Ono (Institute of Science Tokyo): C1_10_9 (40,282.2), R1_10_10 (47,358.5), R1_10_2 (48,189.8), R1_10_3 (44,644.7), R1_10_4 (42,421.2), R1_10_6 (46,883.6), R1_10_7 (43,925.0), R1_10_8 (42,245.8), R1_10_9 (49,140.6), RC1_10_10 (43,518.8), RC1_10_3 (42,104.5), RC1_10_4 (41,350.3) and RC1_10_6 (44,894.9) using an evolutionary search.
October 22, 2026
by CVRPLib
Forty-six improved solutions reported by Shiyao Bian (Independent Researcher): X-n536-k96 (94,832), XL-n1048-k237 (380,091), XL-n1234-k55 (96,597), XL-n1421-k232 (384,636), XL-n1608-k39 (48,001), XL-n1841-k126 (213,941), XL-n1888-k82 (143,552), XL-n2028-k617 (544,140), XL-n2074-k264 (421,225), XL-n2168-k138 (127,188), XL-n2261-k54 (98,822), XL-n2307-k34 (47,933), XL-n2354-k631 (940,625), XL-n2447-k290 (218,627), XL-n2541-k121 (146,343), XL-n2587-k66 (73,248), XL-n2821-k208 (216,603), XL-n2914-k95 (88,898), XL-n3054-k461 (782,157), XL-n3147-k232 (256,404), XL-n3194-k161 (148,591), XL-n3241-k115 (221,246), XL-n3334-k934 (1,452,315), XL-n3408-k524 (678,002), XL-n3484-k436 (703,314), XL-n3561-k229 (209,180), XL-n3640-k211 (189,501), XL-n3975-k687 (525,555), XL-n4063-k347 (548,288), XL-n4153-k291 (355,768), XL-n4340-k148 (243,863), XL-n4635-k790 (609,684), XL-n4738-k487 (759,604), XL-n4844-k321 (404,411), XL-n4951-k203 (284,736), XL-n5406-k783 (1,040,106), XL-n5526-k553 (336,837), XL-n5649-k401 (644,551), XL-n6588-k473 (333,810), XL-n6734-k330 (447,452), XL-n6884-k148 (181,582), XL-n7037-k38 (70,754), XL-n7516-k859 (572,940), XL-n7683-k602 (701,129), XL-n8960-k634 (772,867) and XL-n9160-k379 (323,566) using an intelligent decompose-and-intensify matheuristic built around HGS-CVRP.
October 21, 2026
by Rafael Martinelli
One improved solution reported by Chris Pawlowski (Dayjob.ai): XL-n10001-k1570 (2,329,810).
October 19, 2026
by Rafael Martinelli
One improved solution reported by Chris Pawlowski (Dayjob.ai): XL-n10001-k1570 (2,330,176).
October 13, 2026
by Rafael Martinelli
One improved solution reported by Yuchen Shi (Tongji University): X-n536-k96 (94,834).
October 12, 2026
by Rafael Martinelli
Ten improved solutions reported by Martín Vizzolini (Independent Researcher): Flanders2 (4,364,492), X-n979-k58 (118,924), XL-n2681-k540 (797,844), XL-n6168-k1922 (1,527,163), XL-n8028-k294 (266,255), XL-n8389-k2028 (3,352,860), XL-n9363-k209 (205,365), XL-n9571-k55 (106,341), XL-n9784-k2774 (4,076,860) and XL-n10001-k1570 (2,330,257), by starting with the published BKSs and improved on top of them. Only the solution for XL-n10001-k1570 was built independently.