BARON may not give global optimum although the model status is "1 Optimal"

It seems that BARON may not give global optimum although the model status is “1 Optimal”. Is it?

When I solve my MINLP model with BARON (to maximize), it gives an OBJECTIVE VALUE 85461884.9974 and the MODEL STATUS is 1 Optimal. But, if I fix some of the 0-1 variables at the values BARON gives and solve the same model with BARON again, the obtained OBJECTIVE VALUE changes to 85461884.5701 although the MODEL STATUS is still 1 Optimal. Why?

I would say that numerical error is a likely cause, given that:

( 85461884.9974 - 85461884.5701 ) / (85461884.9974 ) ~= 1 E-9 .

The solver tolerances may very well be higher than this value.