mechanistically, I expect this may be similar to #25515, just instead of dpi, would be theme... not quite sure how to hook into the browser, but conceptually it is similar, at least.

I suspect that there are plenty of plots in our own docs that would not want it (e.g. when explicitly selecting colors that would not work on a dark background), but honestly many of them (if they are just using default color cycle/text colors), it may be better.

I would expect to have a global toggle as well as an individual plot directive toggle.

Update: I finally got around to try using https://github.com/scipy/scipy/blob/c1b803b66a3e9f0416f3feafc6d4cf24d2ec98e2/doc/source/_static/scipy.css#L133 in astropy but it didn't work.

So I had to revert that change from astropy/astropy#15783

Does matplotlib have a native ability to make a dark mode version of a plot or would that setup have to be part of every plot directive codeblock?

The way Furo handles this is it has only-light and only-dark css classes for images, which show/hide the image for the correct mode https://pradyunsg.me/furo/reference/images/#different-images-for-dark-light-mode. Do other themes work the same way? It would be nice if there could be a universal theme independent solution to this that just worked automatically.

I think this is solvable now by setting a custom template + custom pre_code that generates two plots and puts the appropriate classes on them. But that's not a good long term solution because the template is pretty complex and if you use a custom one, it would not stay updated with any upstream updates to the extension.

Another option would be for matplotlib to generate SVGs that do the right thing in dark mode. Is that something that it already supports? See sympy/sympy#25702 for examples of how that can work.