We consider several variants of Keisler’s isomorphism theorem. We separate these variants by showing implications between them and cardinal invariants hypotheses. We characterize saturation hypotheses that are stronger than Keisler’s theorem with respect to models of size $\aleph _1$ and $\aleph _0$ by $\mathrm {CH}$ and $\operatorname {cov}(\mathsf {meager}) = \mathfrak {c} \land 2^{<\mathfrak {c}} = \mathfrak {c}$ respectively. We prove that Keisler’s theorem for models of size $\aleph _1$ and $\aleph _0$ implies $\mathfrak {b} = \aleph _1$ and $\operatorname {cov}(\mathsf {null}) \le \mathfrak {d}$ respectively. As a consequence, Keisler’s theorem for models of size $\aleph _0$ fails in the random model. We also show that for Keisler’s theorem for models of size $\aleph _1$ to hold it is not necessary that $\operatorname {cov}(\mathsf {meager})$ equals $\mathfrak {c}$.