The Free branch model that lets ω being estimated for each branch would be the best option in this case.

A good option to see how are varying evolutionary rates across codons is the M2 model, in particular once run have a look to the Bayes Empirical Bayes (BEB) probabilities in the rst outfile. But it is important to note that those are descriptive values, if you want to know if some sites are under positive selection you will need to do a likelihood ratio test between M1 (neutral model) and M2. Only if M2 has significantly a better fit, sites with ω > 1 will be considered to be under positive selection.

To detect if branch B is under positive selection you will need to run 2 branch models. A first one fixing ω = 1 for your branch B and the second letting it free. Than if under letting branch B free its ω is higher than 1 and if free branch model has a better fit than fixed branch, branch B will be under positive selection.