Previous research suggests that individuals who play action video games tend to outperform non-action video game players in various cognitive functions. In the current study, we compare the precision of visual short-term memory for the direction of coherent motion among action video game players (AVGPs) and age and gender-matched non-action video game players (NAVGPs). Participants were tasked with remembering the direction of up to four random dot kinematograms (RDKs) presented in a sequence, with the number of RDKs varying randomly between one and four in each trial. After the sequence, a probe number appeared on the screen, indicating which RDK the participant needed to reproduce. Following either a brief (0.5 s) or a longer interval (3 s), participants reproduced the direction of a designated RDK. To reach a specific performance level with a single RDK stimulus, participants underwent initial training. The training results revealed that AVGPs required slightly fewer training blocks than NAVGPs to reach the desired performance level. While the two groups did not differ in precision for the short delay, AVGPs performed the task more precisely than NAVGPs when tasked with reproducing the motion direction of a single RDK after a long interval. Using a complementary model-fitting methodology, we investigated the distribution of memory resources within the two groups. Our analysis revealed that, in contrast to the AVGPs, NAVGPs exhibited higher neural costs relative to the behavioral advantages linked to memory encoding. This was particularly evident during the long retention interval, resulting in NAVGPs allocating a reduced amount of memory-related resources in this condition. Future training studies could explore the integration of models that account for the allocation of memory resources, balancing the behavioral benefits of encoding precision against the neural costs involved.