The aim of this study was to compare internal and external load profiles of different game profile-based training (GPBT) formats among elite young football players. Twenty-one participants (age: 18.7 ± 0.6 years) performed three sessions of three GPBT formats, which were matched for training volume but structured with different high-speed running and sprint demands: i) performed along linear paths (GPBT-L); ii) performed as repetitive actions of short distance including many multi-directional changes of direction (GPBT-S) and, iii) a combination of the other two protocols, that is linear high-speed runs and sprint efforts with a single change of direction (GPBT-M). External load outputs were collected using GPS units, physiological and perceptual responses were monitored with heart rate (HR) monitors, and ratings of perceived exertion (RPE), respectively. While no differences were found between formats for HR and RPE, distinct external load profiles were observed for high-speed running (HSD) and sprint distances (SD), (GPBT-L > GPBT-M > GBPT-S, all p < 0.05), and high-intensity acceleration and deceleration efforts (HIE), (GPBT-S > GPBT-M > GPBT-L, all p < 0.05). Moreover, the GPBT-S format was characterized by greater intra-session variability for HSD, SD, and HIE (CV% = 24.2%, 16.5% and 20.4%, respectively) and inter-session variability for HSD and SD (CV% = 10% and 15.7%, respectively) compared to the other two formats. Considering their load profiles and the associated reliability scores, football practitioners can implement GPBT formats interchangeably to elicit necessary internal load responses and selectively to prioritize specific external load outputs.