Musical training (MT) is perceived as a multi-sensory program that simultaneously integrates visual, aural, oral, and kinesthetic senses. Furthermore, MT stimulates cognitive functions in a ludic way instead of tapping straight into the traditional context of school learning, including mathematics. Nevertheless, the efficacy of MT over mathematics remains understudied, especially concerning longstanding effects. For this reason, this longitudinal study explored the impact of MT on numerical cognition and abstract visual reasoning using a double-blind and quasi-experimental design. We assessed two groups of children from primary schools, namely one with developmental dyscalculia [DD; n = 22] and another comprising typically developing children [TD; n = 22], who concomitantly underwent MT. Numerical cognition measurement was carried out at four different time points: Baseline (pre-MT assessment), mid-test (after 7 weeks of MT), post-test (after 14 weeks of MT), and follow-up (10 weeks after the end of MT). Significant interactions were found between time and group for numerical cognition performance, in which the DD group showed higher scores in number comprehension, number production at mid-test, and calculation at post-test compared to baseline. A key finding was that number production, number comprehension, and calculation effects were time-resistant for the DD group since changes remained on follow-up. Moreover, no significant differences over time were found for abstract visual reasoning for both groups. In conclusion, the findings of this study showed that MT appears to be a useful tool for compensatory remediation of DD.