Clickbaity title, since they don't have the answer and the paper says as much. They do have some guesses, anyone interested just skip the article and go to the conclusions section.
Guess 1 is female student behavior adaptation to what teachers want aside from academic performance:
> School and classroom environments might indeed be adapted to traditionally female behaviours (Lavy 2008). Female students might thus adopt such actual behaviours during class, including precision, order, modesty, and quietness, which go beyond the individuals’ academic performance, but which teachers may highly reward in terms of grades. ... Conversely, teachers may be likely to associate such behaviours only with female students, because girls are traditionally thought of as possessing these traits.
Guess 2 is gender bias over-compensation:
> Another theoretical explanation calls into play teacher overcompensation towards females. Girls are indeed often discussed in discrimination contexts, especially in speeches about gender differences in cognitive ability in dealing with the ‘hard subjects’. A possible explanation for the reason teachers are more generous in grading female students could be that teachers wish to avoid possible discrimination against girls as an ability-stigmatized group. Therefore, teachers may over-assess girls in the same way they sometimes over-assess non-native students, to avoid negative stereotyping (Alesina 2018). From another perspective, it is also possible that the over-assessment of female students’ competences in Mathematics partly represents a sort of ‘push’ to encourage the weaker students (Terrier 2015). As a consequence, this may translate into a positive discrimination which favours girls.
Guess 3 is simply noisy data (teachers using grades to manipulate students into performing better, according to whatever teachers think motivates boys vs. girls):
> A final interpretation of our results could stem from a limitation of our data. Indeed, our measure of student grading is derived from the midterm report card, rather than the final report card of the academic year. Teachers may consciously adopt specific grading practices at the midterm to trigger students’ effort differently according to their gender, to in turn obtain the best possible performance by the end of the academic year. Specifically, teachers could perceive that male students’ effort is more easily triggered by lower grades, which could encourage them to study harder to achieve a better grade, while female students’ effort to achieve higher grades may be more easily triggered via encouragement.