Despite the growing interest in game-based learning (GBL) over the last two decades, the influence of Augmented Reality-based GBL on English vocabulary growth is still not well understood when compared with other forms of media, especially with regard to academic English vocabulary. This quasi-experimental study with 90 Chinese English-as-a-foreign-language (EFL) university students investigated the efficacy of GBL supported by three forms of media (paper, digital, and AR) for learners' academic English vocabulary growth by measuring their immediate recall and long-term retention. The study involved one experimental and two control groups, each learning 17 target words relating to the digestive system by using AR-based GBL (AR-GBL), digital-based GBL (D-GBL), and paper-based GBL (P-GBL) approaches. Participants completed a pretest, an immediate posttest, and two delayed posttests (one and three weeks after the intervention) on these target words. A mixed 3 × 4 repeated measures analysis revealed a quadratic trend in vocabulary growth across all three media-based GBL approaches. AR-GBL's effect on immediate academic vocabulary learning was superior to D-BGL's and equivalent to P-GBL's effect. All approaches exhibited similar declines in delayed retention after one week, with no significant differences. Notably, AR-GBL indicated a marginally significant difference compared to D-GBL in terms of retention at the three-week delayed posttest. These findings underscore the need to capitalize on each approach's unique benefits as well as spaced learning to optimize EFL academic vocabulary learning.