$PA$ is a weird theory to work with, here: if you want to talk about models, then a two-sorted theory like $RCA_0$ or $ACA_0$ is probably better. $PA$ cannot directly talk about models, since it can't directly talk about sets.
However, $PA$ (in fact, much less than $PA$) proves that if $PA$ is consistent, then so is the theory $PA^*$ in the language of arithmetic + one new constant, $c$, and containing the axioms $c\not=n$ for each (standard) natural number $n$; so in that sense, $PA+Con(PA)$ would prove that $PA$ has nonstandard models.
If one wants to talk about models directly, using a two-sorted theory, then things get more interesting: there is a reasonable theory, $RCA_0$ - its first-order part is $I\Sigma_1$, $PA$ but with induction substantially restricted - which can prove that $Con(PA)$ implies $Con(PA+\neg Con(PA))$, and which proves that complete consistent theories have models, but can't quite prove that arbitrary consistent theories have models. So, for example, $RCA_0+Con(RCA_0)$ cannot prove that $RCA_0$ has nonstandard models (the easiest way to see this is via Tennenbaum's theorem).
Note that this added complexity means that the distinction between models and consistencies [sic] is actually very significant, and not something to be lightly swept under the rug.