tl;dr: There could be multiple explanations depending on how you want to treat Wikipedia. If you want to treat Wikipedia as public good where everyone contributes a small part towards its creation and that everyone then enjoys equality you can explain it as people trying to still satisfy their own preferences through consuming the final Wikipedia page.
You could also treat it as an example of pure altruism people provide for others because they enjoy knowing that others are now better as a result. Furthermore, it could also be treated as completely self-interested action because one enjoys it as a hobby or because they derive some other personal benefit such as improving their own skill as mentioned in the answer provided by Steve. It is also possible it is a mixture of these possible explanations.
Voluntary Provision of Public Goods:
There are actually several models of voluntary public good provision, these are nowadays even included in textbooks (see Mueller Public Choice III for example). For example, we could model the Wikipedia situation as an adapted version of one of the voluntary public provision goods models presented by Mueller in his book:
For example, let's subdivide Wikipedia into sub-Wikipedias by for example its subjects and model contribution to single subject area at a time (as suggest by Michael in his +1 comment). The total contribution to that particular subject area on Wikipedia will be our public good $W$ that will consist as a collection of individual article contributions to the subject $W_i$ so $W=W_1+W_2+...+W_n$. Individual utility will be given as $U_i(x_i,W)$ where $x_i$ is some standard consumption and $W$ is the Wikipedia a public good which is consumed by everyone. An individual budget constraint of individual will be given by $M = P_xx_i + P_w W_i $ where $M_i$ is a budget $P_x$ is the price for consumption and $P_w$ 'price' for the individual's contribution to Wikipedia - this is an abstraction of course in real life one does not pay a price to 'purchase' and post Wikipedia contribution but it is just simple way how to avoid explicitly modeling everything in terms of labor supply vs leisure trade-off and will save me a lot of work and make this problem shorter without any substantial change in result. Hence under the above assumption an individual optimum choices would be given by solving the following Lagrangian:
$$L = U_i(x_i,W) - \lambda_i(M_i - P_x x_i - P_w W_i)$$
which gives us the following FOC's:
$$\frac{\partial U_i}{\partial W} - \lambda_i P_w =0 $$
and
$$\frac{\partial U_i}{\partial x_i} -\lambda_i P_x=0$$
hence the condition for utility maximization is given by:
$$\frac{\partial U_i/\partial W}{\partial U_i/ \partial x_i}=\frac{P_w}{P_x}$$
from this we can even calculate an individual contribution to Wikipedia to be by specifying some exact utility function. For example, suppose that utility is given by Cobb-Douglas as $U_i= x_i^a W^b$ which would imply that individual contribution to creating Wikipedia in equilibrium will be:
$$W_i = -\frac{a}{a+b} \sum_{j\neq i} W_j + \frac{b}{a+b} \frac{M_i}{P_W}$$.
The above result is extremely intuitive as it shows a free rider behavior, the first term shows that the more other people contribute to Wikipedia the more you will free ride and write less of your own. The second term just shows that the higher your income is or the lower the 'price' for writing is the more you write.
We can even calculate the total contribution the give Wikipedia subject by whole community. For example, assume everyone has the same income $M$ to simplify math as it will imply that everyone will choose exactly the same contribution, then the total contribution of whole community will be given by:
$$W = nW_i = n \left(-\frac{a}{a+b} (n-1) W_i + \frac{b}{a+b} \frac{M}{P_W} \right) = \frac{nb}{an+b} \frac{M}{P_w}$$
Its also worth noting that this contribution will also be lower than pareto-optimal contribution (except for special cases such as a case where marginal utility of $x$ is zero) due to the free riding issue but it wont zero save special cases such as if income would be zero or 'price' for contributing to Wikipedia would tend to infinity in the limit etc.
This model would provide valuable insight especially if we can argue that Wikipedia entries are not solely written for the benefit of readers but also the writers themselves as pointed by Giskard in his insightful comments or argued implicitly by Steve.
Moreover, as per Michael's valuable suggestion this model is more appropriate when applied to individual sub-Wikipedias. Hence these would be better way to model contributions to for example different subject-matters, but at the same time Wikipedia is ultimately a collection of all individual sub-Wikipedias.
Altruistic Explanation:
In case we would not want to assume that writer derives any benefit from writing Wikipedia articles then we can treat more as an charitable activity. In that case you could explain it by utility function that is dependent also on utility of others. For example, you could model that person's utility as being interdependent. For example, with utility $U_i=\Psi( u_i(x), u_j(x))$ where $\partial U_i/ \partial u_j >0$ (see for example Hori 2002 for an more complex example). In this case people would provide Wikipedia article just because they would enjoy knowing that some other people will derive some utility/benefit from reading the articles.
Other explanations:
Alternatively as mentioned by Steve you could treat it as a learning experience where individuals write these articles as investment in improving their human capital. In that case you can view them as cost of accruing more human capital which will bring benefits later in form of higher income thanks to having better communication or retention skills (see an overview of theory of human capital in Becker, Gary S. Human capital: A theoretical and empirical analysis, with special reference to education).
You could also view it as a 'hobby' so it would be work one is doing for themselves where utility is derived actually from performing the task and I bet that some other explanations might exist as well.
To what extent would economists consider this a "puzzle"?:
This is a good question but quite subjective one. There are some authors that refer to voluntary provision of public goods as a 'puzzle' (see Anderoni 1995), but I also don't think that this ever became puzzle in the same prominent way as let's say equity-premium puzzle. Also I don't think contemporary public economists still consider it to be an unsolved puzzle - there are some related puzzles where some experiments show that sometimes people tend to contribute more than expected - but there are also all sort of behavioral/evolutionary/repeated game explanations for that.