More Contrast Generally Better
There are no general standards for capping contrast for either day or night displays. Generally, the higher the contrast, the easier the reading in both cases (Parker & Schaff, 1998; Zuffi, Brambilla, Beretta, & Scala, 2007). Thus, your UI can have two independent dimensions: day versus night, and low versus high contrast, where high contrast is WCAG AAA compliant whether it’s day or night mode.
I would treat user complaints about excessive contrast with caution. Given that lower contrast (dark gray font) has recently been fashionable on many web sites, user may prefer it more because it looks cool and modern, rather than because it’s actually easier to read.
Contrast and Dyslexia
The only concerns about high contrast I’ve heard (and really the only reason to have a low-contrast mode at all) is that some individuals with dyslexia report letters appear “jumpy” at high contrast settings (Rello, Kanvinde, & Baeza-Yates, 2011). However, the effect is not well established or understood (e.g., O’Brien, Mansfield, & Legge, 2000), and there appears to be considerable individual variation (Rello, Kanvinde, & Baeza-Yates, 2011). It may be best to let the user choose the contrast, maybe continuously with a slider. In your comment exchange with JonW, you are right that user can pick horrible color combinations, but if you limit the range of the slider to between WCAG AAA and AA levels, users can’t make it too bad, and, in fact, users with disabilities seem to do okay with choosing their contrast and even (subtle) colors (Gregor & Newell, 2000).
So, I recommend making contrast user-adjustable within the limits of WCAG, but to default to high contrast, and perhaps provide some in-line documentation on how to adjust the contrast slider (“Accessibility: slide until letters are easiest to see and don’t seem to move”).
Maybe the Problem is Night Mode
Now, about night mode. You don’t describe your environment of use, but for everyone’s benefit, light-on-dark font does not make text easier to read at night. Dark-on-light is consistently easier to read in light or dark conditions (Buchner & Baumgartener, 2007; Dobres, Chahine, & Reimer, 2017). There are several theories for this (Scharff & Ahumada, 2003), but it might come down to modern computer fonts being designed for dark-on-light use. In any case, your user complaints about the light-on-dark being too contrasty might really be complaints about light-on-dark itself being hard to read, and not really a problem with the contrast.
The purpose of night mode is to make everything else at night easier to see. That is, by reducing the amount of light shining from the device, the user maintains dark adaptation in order to see other things. If your users are not interacting with dark objects while using your device, then you don’t want light-on-dark regardless of the ambient illumination.
Even if your users do need to maintain dark adaption, you may want to consider other methods to reduce the light coming from the display. For example, you may want to physically dim the display or even put a physical neutral density filter over it, all while keeping the easier-to-read dark-on-light text. Keep in mind, however, that anything you do to reduce display illumination will reduce display readability. As the human eye dilates for dark adaptation, focusing becomes more difficult, especially for older users. You need to decide if it's worth the trade-off in your case.
Buchner & Baumgartner, 2007. Text – background polarity affects performance irrespective of ambient illumination and colour contrast. Ergonomics, 50, 1036-1063.
Dobres, Chahine, & Reimer, 2017. Effects of ambient llumination, contrast polarity, and letter size on text legibility under glance-like reading. Applied Ergonomics, 60, 68-73.
Gregor & Newell, 2000. An Empirical Investigation of Ways in Which Some of the Problems Encountered by Some Dyslexics May be Alleviated using Computer Techniques. Proceedings of the fourth international ACM conference on Assistive technologies, Arlington, VA.
O’Brien, Mansfield, & Legge, 2000. The effect of contrast on reading speed in dyslexia. Vision Research 40, 1921–193.
Parker &, Scharff, 1998. Influences of Contrast Sensitivity on Text Readability in the Context of a Graphical User Interface.
Rello, Kanvinde, & Baeza-Yates, 2011. Layout Guidelines for Web Text and a Web Service to Improve Accessibility for Dyslexics. Proceedings of the International Cross-Disciplinary Conference on Web Accessibility, Lyon, France.
Scharff & Ahumada 2003. Letter identification performance is better for negative contrast than positive contrast. Vision Sciences Society Annual Meeting, Sarasota FL
Zuffi, Brambilla, Beretta, & Scala, 2007. Human Computer Interaction: Legibility and Contrast. 14th International Conference on Image Analysis and Processing (ICIAP 2007).