So suppose $\kappa$ inaccessible so that $V_\kappa$ is a model of ZFC, using Skolem and the Mostoswky collapse we have a countable elementary submodel $M$ of $V_\kappa$. This implies that for any formula $F(x)$ and any $a$ in $M$ we have : $V_\kappa$ verifies $F(a)$ iff $M$ verifies $F(a)$.
$\omega$ (the set of all integers) is absolute between transitive models so the formula "there is a bijective $f$ : $\omega$ into $A$" has only $A$ as a free variable.
Suppose $A$ is countable (and so countable in $V_\kappa$) we have $V_\kappa$ verifies : "there is a bijective $f$ : $\omega$ into $A$"
If $A$ is in $M$, by $M$ being an elementary submodel we have $M$ verifies : "there is a bijective $f$ : $\omega$ into $A$", so $A$ is countable in $M$, this implies that $M$ thinks that every (not finite) set is countable, which is false.
I can't find were the error is, I know that $M$ is supposed to not have such $f$ so that $M$ thinks that some sets are uncountable.