HiGHS generates a different result from other solvers

[zxt5]

HiGHS returns different results from gurobi, scip and GLPK on seed.mps.txt, but actually is the same as the answer of CBC.

Solver: HiGHS, CBC Status: Optimal Objective: 2363.670185116735 x0 = -17.0, x1 = -1.9234801367308, x10 = 2.0, x11 = -17.0, x12 = 12.0, x13 = 8.0, x14 = 16.0, x15 = -6.0, x2 = -2.3885052868101, x3 = -6.3174490195748, x4 = -13.0, x5 = 14.4843573578535, x6 = 15.0, x7 = 3.3895579071107, x8 = -2.6028842710044, x9 = -6.0,

Solver: Gurobi, scip, GLPK Status: Optimal Objective: 2294.7109491954575 x0 = -15.0, x1 = -1.4332546678040998, x10 = 2.0000030474431907, x11 = -17.0, x12 = 13.0, x13 = 9.0, x14 = 16.0, x15 = -6.0, x2 = -2.645833997488699, x3 = -7.8773257828424015, x4 = -14.0, x5 = 13.87545430453603, x6 = 14.0, x7 = 3.2154789890426607, x8 = -2.245099829481236, x9 = -6.0,

Not sure if this is tolerable.

[jajhall]

This is a relative difference of 0.03, so well outside anything that could be explained by the relative MIP gap tolerance in HiGHS.

[zxt5]

I agree, and setting presolve on/off doesn't change anything.

[jajhall]

You've generated a remarkable number of issues recently. What proportion of your instances do they represent?

[zxt5]

The proportion is not large, given the large number of all the instances generated.

I am actually using some generation approaches to generate a large number of cases for stress testing different solvers.