Prior research shows that gender differences in self-efficacy and test anxiety can predict students' course performance. Although there has been research that uses test anxiety and self-efficacy to predict student grades, no study to our knowledge has investigated this in the context of low- and high-stakes physics assessments. In this work, we used survey data and grade information to compare the predictive power of self-reported self-efficacy and test anxiety on low-stakes (e.g., homework and quizzes) and high-stakes (e.g., traditional exams) assessment scores in a traditionally taught large introductory physics course. We found that there are gender differences in both self-efficacy and test anxiety, as well as in high-stakes assessment outcomes. There were no gender differences in low-stakes assessment scores. Further, we found that models that control for self-efficacy and/or test anxiety eliminate the predictive power of gender for high-stakes assessment scores. Finally, we found that self-efficacy partially mediates the effect of test anxiety on high-stakes assessment outcomes. These trends suggest that teaching methods that alleviate test anxiety have the potential to create a more equitable learning environment.