Currently we are fitting a generalized linear model to a set of data using R on Fedora 33. We are 3 people that have the exact same data set, exact same OS (Fedora 33) , exact same R version and package version. The sessionInfo() output also looks the same. The parameters for glm are also the same and in particular we have 'family = binomial(link = "probit")' for all of them.
Something to note is that a normal "lm" fit yields the exact same results. Something about "glm" yields different results on different computers (by a tiny amount). Finally, the results are consistent in the sense that we all three get the same result with the same data in the same computer, only when we compare results is when the problem appears.