I know what the z-value of a single observation is, that is explained in Wikipedia. But what is the z-value of a parameter in a glm model?
It is the test-statistic for the Wald-test that the parameter is 0. It is the parameter divided by the standard error. If the null-hypothesis is true (i.e. the parameter is 0 in the population) and we were to draw many samples from our population and compute the z-statistic in each of them, then those z-statistics will follow a standard normal distribution.