If there is a transitive set model $M$ of ZFC, then all finite successor ordinals $n+1$ and all $\omega+n+1$ are $\upsilon(M)$ for some such $M$, and also many other ordinals:
Claim: Let $\Omega,\alpha<\omega_1^L$ be such that $L_\Omega\models\mathrm{ZFC}$ and $0<\alpha$ and there is a surjection $f:\omega\to\Omega+\alpha$ in $L_{\Omega+\alpha+1}$. Then there is a transitive model $M\models\mathrm{ZFC}$ of height $\Omega$ such that $\upsilon(M)=\alpha+1$; moreover, in $L_{\alpha+1}(M)$ there is a surjection $g:\omega\to M$.
Remark: Note there calso be $\beta<\alpha$ such that $L_{\Omega+\beta+1}$ has a surjection $f':\omega\to\Omega+\beta$, and hence a surjection $f'':\omega\to\Omega=\mathrm{OR}\cap M$, but there won't be any wellorder of $M$ in $\mathcal{J}_{\Omega+\beta+1}$. The claim gives in particular that if there is a transitive $M\models\mathrm{ZFC}$
then all finite ordinals, $\omega+1$, and many "small" successor ordinals are $\upsilon(M)$ for some $M$. (E.g. if $\Omega$ is least such that $L_\Omega\models\mathrm{ZFC}$ then for each of these "small" ordinals $\alpha$ we can find an $M$ of height $\Omega$ with $\upsilon(M)=\alpha+1$, and also for cofinally many $\alpha<\omega_1^L$ there is $M$ of height $\Omega$ with $\upsilon(M)=\alpha+1$. On the other hand, if there is say a transitive model $M$ of ZFC of cardinal height $\kappa$, then $L_\kappa$ also models ZFC, and for whatever ordinal $\alpha>0$, we get that $L_{\kappa+\alpha}\models$``$\kappa$ is a cardinal'',
and by taking the $\Sigma_{n+1}$-hull in $L_{\kappa+\alpha}$ of the single parameter $\kappa$, we get $L_{\bar{\kappa}+\bar{\alpha}}$ modelling the same about $\bar{\kappa}$, and $L_{\bar{\kappa}}\models\mathrm{ZFC}$, and there is a surjection $f:\omega\to\bar{\kappa}+\bar{\alpha}$ in $L_{\bar{\kappa}+\bar{\alpha}+1}$.)
Proof: I'm going to work with the $\mathcal{J}$-hierarchy, not the $L$-hierarchy, since particularly for dealing with successor levels, the $\mathcal{J}$-hierarchy is smoother. This doesn't affect the outcome, because we will have $\mathcal{P}(M)\cap\mathcal{J}_\alpha(M)=\mathcal{P}(M)\cap L_\alpha(M)$ for each ordinal $\alpha$, and hence they agree about when a wellorder of $M$ is constructed. (It is easy enough to see that in fact $L_\alpha(M)\subseteq \mathcal{J}_\alpha(M)$ for each $\alpha$; for the converse, one must code $\mathcal{J}_\alpha(M)$ over $L_\alpha(M)$, uniformly enough in $\alpha$.) If one prefers to work only with the $L$-hierarchy, one should be able to translate everything that follows into it (swapping $\mathcal{J}_\beta$ for $L_\beta$ and $\mathcal{J}_\beta(M)$ for $L_\beta(M)$), but then the proof would directly involve more coding work.
Recall that if $\beta$ is sufficiently closed then $\mathcal{J}_\beta=L_\beta$; this holds in particular for $\Omega$, since $L_\Omega\models\mathrm{ZFC}$.
The plan is to obtain $M$ as a class generic extension of $\mathcal{J}_\Omega$, via generic $G$, setting $M=\bigcup_{\alpha<\Omega}\mathcal{J}_\Omega[G\upharpoonright\alpha]$ (in particular, $G$ itself is not included as a predicate of $M$), and with enough symmetry that by taking $G$ generic over $\mathcal{J}_{\Omega+\alpha}$, $\mathcal{J}_\alpha(M)$ cannot have a wellorder of $M$; but using that $\mathcal{J}_{\Omega+\alpha+1}$ has a surjection $f:\omega\to \mathcal{J}_{\Omega+\alpha}$, we will take $G\in \mathcal{J}_{\Omega+\alpha+1}$, and we therefore get a surjection $g:\omega\to M$ in $\mathcal{J}_{\alpha+1}(M)$. It will be similar to the standard construction of models of ZF + $\neg\mathrm{AC}$ of the form $\mathrm{HOD}(\mathbb{R}^*)^{V[G]}$, for some reasonably symmetric set of reals $\mathbb{R}^*$, and some generic $G$ for some homogeneous forcing. But it is finer in detail, since we are aiming for choice to fail right up to some point, and then to hold right after that.
So, work for the moment in $\mathcal{J}_\Omega$; we specify the forcing. For $\kappa$ regular let $\mathbb{P}_\kappa$ be the forcing to add a subset of $\kappa$ via conditions $f:\beta\to 2$ for $\beta<\kappa$, and let $\mathbb{P}$ be the class product of all such $\mathbb{P}_\kappa$, with set-sized supports in the indices $\kappa$. (That is, conditions are functions $f:R\to V$ where $R$ is a set of regular cardinals and $f(\kappa)\in\mathbb{P}_\kappa$ for each $\kappa\in R$, and the ordering is by extension.)
Note that for each regular $\gamma$, we can factor $\mathbb{P}$ as $(\mathbb{P}\upharpoonright\gamma)\times(\mathbb{P}\upharpoonright[\gamma,\infty))$, and $\mathbb{P}\upharpoonright[\gamma,\infty)$ is ${<\gamma}$-closed.
Using the enumeration $f:\omega\to \mathcal{J}_{\Omega+\alpha}$ in $\mathcal{J}_{\Omega+\alpha+1}$,
let $G\in \mathcal{J}_{\Omega+\alpha+1}$ be such that $G$ is $(\mathcal{J}_{\Omega+\alpha},\mathbb{P})$-generic (here I mean generic for all dense sets which
are elements of $\mathcal{J}_{\Omega+\alpha}$; recall $\alpha>0$, so this includes all dense sets which are definable from parameters over $\mathcal{J}_\Omega$.)
Let $M$ be defined as indicated above. Then $M\models\mathrm{ZFC}$.
Because we have a surjection $h:\omega\to L_{\Omega}$ with $h\in\mathcal{J}_{\Omega+\alpha+1}$,
and $G\in\mathcal{J}_{\Omega+\alpha+1}$, we get a surjection $g:\omega\to M$ with $g\in \mathcal{J}_{\Omega+\alpha+1}\subseteq \mathcal{J}_{\alpha+1}(M)$.
So we just need to see there is no wellorder of $M$ in $\mathcal{J}_\alpha(M)$. This takes some work regarding the forcing details, but other than that, it is like in the constructions of models of $\neg\mathrm{AC}$ like mentioned above.
One first needs to work through the forcing details, to see level-by-level that $\mathcal{J}_{\Omega+\beta}$ has names for all elements of $\mathcal{J}_\beta(M)$, and the forcing relation is level-by-level and quantifier-by-quantifier appropriately definable. (Such level-by-level forcing calculations are commonly used in inner model theory, so there are some papers where they can be found). Here is a sketch of what we need in this case. At the level of $\mathcal{J}_\Omega$ and $M$ itself, note that all elements of $M$ have names in $\mathcal{J}_\Omega$, and the $\Sigma_n$-forcing relation is $\Sigma_{n+k}$, for some fixed $k$, uniformly in $n$. Then proceed inductively on $\beta>0$. Define canonical names for elements of $\mathcal{J}_\beta(M)$, corresponding to how things are defined in the $\mathcal{J}$-hierarchy; these are natural kinds of symmetric names. E.g. the subsets of $M$ which are in $\mathcal{J}_1(M)$ are just those which are definable from parameters over $M$, so we can take names for such sets as being tuples $\sigma=(\varphi,\tau)$ where $\tau\in\mathcal{J}_\Omega$ is a $\mathbb{P}$-name and $\varphi$ is a formula, and then this name gets interpreted as $\sigma_G=\{y\in M\bigm|M\models\varphi(y,\tau_G)\}$. These canonical names should, moreover, each have a certain support in $\mathcal{J}_\Omega$, corresponding to some bound on the parameters in $M$ needed to define the object in the generic extension $\mathcal{J}_\beta(M)$. That is, if a set $A\subseteq M$ is defined over $M$ from $x\in M$, then we can say that $\gamma$ is the support of $A$, where $\gamma$ is least with $x\in V_\gamma^M$. Higher up, if $A\in\mathcal{J}_{\beta+1}(M)$ is defined over $\mathcal{J}_\beta(M)$ from parameters $x_1,\ldots,x_n\in\mathcal{J}_\beta(M)$, then these $x_i$'s already have some supports $\gamma_i<\Omega$, then their supremum gives a support for $A$. The canonical names are just codes for these kinds of definitions of objects from finitely many ordinals and elements of $M$, and hence can also be assigned some support $<\Omega$. For $\beta>0$, the $\Sigma_n$-forcing relation over $\mathcal{J}_{\Omega+\beta}$ (for $\Sigma_n$ facts about $\mathcal{J}_\beta(M)$) will be $\Sigma_{n+k}$-definable over $\mathcal{J}_{\Omega+\beta}$ in the parameter $\Omega$, with some fixed $k$, uniformly in $\beta$. (Actually you can state it much more precisely than this, but one has to think a bit about what exactly the "$\Sigma_n$ forcing relation" should mean.))
Now suppose for a contradiction that we have a wellorder $<^*$ of $M$ in $\mathcal{J}_\alpha(M)$. Then we get a canonical name $\tau\in\mathcal{J}_{\Omega+\alpha}$ such that $\tau_G={<^*}$ (and here note the canonical name $\tau$ gets interpreted naturally in the $\mathcal{J}_\beta(M)$ hierarchy). Since $M\models\mathrm{ZFC}$, from $<^*$, it is easy to recover a surjection $g:\Omega\to M$ (consider the $<^*$-least wellorder of $V_\alpha^M$, for each $\alpha<\Omega$, and use these wellorders (which are each in $M$) to define $g$). So we get such a $g\in\mathcal{J}_\alpha(M)$. Let $\gamma<\Omega$ be a support for $g$. Then $g$ is defined over some $\mathcal{J}_\beta(M)$, where $\beta<\alpha$, from ordinal parameters and parameters in $V_\gamma^M\subseteq \mathcal{J}_\Omega[G\upharpoonright\xi]$ for some $\xi<\Omega$. Let $X=G_\kappa$ be the generic subset of some $\mathcal{J}_\Omega$-regular $\kappa\geq\xi$. Then $X\in M$ but $X\notin\mathcal{J}_{\Omega+\alpha}[G\upharpoonright\xi]=\mathcal{J}_\alpha(\mathcal{J}_\Omega[G\upharpoonright\xi])$, since $G$ is $\mathcal{J}_{\Omega+\alpha}$-generic. Let $\zeta<\Omega$ be such that $X=g(\zeta)$. Now noting that $\mathbb{P}\upharpoonright[\xi,\Omega)$ is sufficiently homogeneous, we see that $X$ is actually definable over $\mathcal{J}_{\Omega+\beta}[G\upharpoonright\xi]$, hence $X\in\mathcal{J}_{\Omega+\alpha}[G\upharpoonright\xi]$, a contradiction.
