Let $\overline{M}_{g,n}(X,\beta)$ be the moduli of stable maps into $X$ of class $\beta \in H_2(X)$. We have the evaluation maps $\operatorname{ev}_i : \overline{M}\_{g,n}(X,\beta) \to X$. Given $\alpha_i \in H^\ast(X)$, the *Gromov-Witten invariant* corresponding to the tuple $(X,\beta,g,n,\alpha_i)$ is the integral $$\int_{[\overline{M}_{g,n}(X,\beta)]^\text{vir}}\bigwedge_i \operatorname{ev}_i^\ast(\alpha_i).$$

There is also the "forgetful" map (or "stabilization" map) $F : \overline{M}_{g,n}(X,\beta) \to \overline{M}\_{g,n}$. I don't know if this is the standard terminology (is it?), but one can define the *Gromov-Witten class* corresponding to the tuple $(X,\beta,g,n,\alpha_i)$ to be the (virtual) pushforward $$F_\ast^\text{vir}\left(\bigwedge_i \operatorname{ev}_i^\ast(\alpha_i)\right) \in H^\ast(\overline{M}_{g,n}).$$

Question: Are there any nontrivial cases in which these Gromov-Witten classes have been identified explicitly, e.g., in terms of say tautological classes on $\overline{M}_{g,n}$?

Just as a very naive motivation, note that in many situations Gromov-Witten *invariants* are zero simply because of "stupid reasons", like degree/dimension reasons (i.e. the degree of the integrand doesn't match the virtual dimension), and so provide no information. But the Gromov-Witten *classes* may still be nonzero and contain some information.