Sampling and Experiments · Superchallenge problem

A randomized experiment puts $4$ subjects in each group. Treatment outcomes: $7, 9, 11, 13$; control outcomes: $3, 5, 6, 8$. Assuming the treatment has no effect (so all $70$ ways of splitting the eight outcomes into two groups of $4$ are equally likely), what is the probability of a treatment-minus-control difference in means at least as large as the observed one?