It is generally understood that girls develop a small to moderate deficit in math abilities, compared to boys, over the course of schooling, as measured by mean school grades or test scores (Hyde & Linn, 2006 give a number of .08 standard deviations in favor of men for mathematical problem solving on average, a larger effect favoring elementary school girls, and no difference for high school math; on the maths section of the SAT, girls score ~.3 SD below boys). Similarly, particularly math-heavy academic disciplines (such as physics, maths, computer science) feature a smaller proportion of women than other disciplines, although this effect is attenuated for some surprising cases, such as statistics, and present for some math-light degrees, such as philosophy. This effect is the stronger, the further advanced the position is (e.g., the ratio of female maths BAs is much higher than the ratio of female math PhDs).
All in all, across these various measures, women tend to score slightly worse than men, although not across the board, and extreme differences are only observable in comparatively rare and extreme contexts (e.g., math PhDs).
The variance ratio (that is, how much more men vary compared to women) is somewhere around 1.1 as given by Hyde, indicating that men show a bit, but not much more variance.
These findings concern data from the United States. Janet Hyde has published multiple cross-national meta-analyses on this topic, including:
Cross-National Patterns of Gender Differences in Mathematics: A Meta-Analysis
Gender, culture, and mathematics performance
Gender Similarities in Mathematics and Science
Her primary finding is the culture dependence of this gender gap. For example, she observes the gap not only varies with countries, but also interacts with, for example, gender equality (although see Fryer & Levitt, 2009). It is different for various ethnic groups within countries (e.g., amongst US Asians, women are somewhat more frequently top scorers). Furthermore, the effects are highly variable in time; for example, the ratio of female math PhDs hit an all-time low of 5% in the 1950s, was higher in the pre-war time, and has been on a steady upwards trend ever since.
Combining all of this evidence, it seems to me fair to say that there is currently, in Western countries in this time, somewhat stronger math performance by boys, but that there is no convincing evidence that this pattern is necessarily generalizable.