This paper explores the relationship between the recently introduced Lion Optimizer and the similar signSGD with momentum --- focusing on their different gradient averaging mechanisms prior to the sign operator. A precise formula for the averaging mechanisms of Lion and Signum is derived, and a comparison is made in the deterministic convex setting. Empirical investigations highlight the advantage of Lion, a finding further supported by a convergence guarantee in the stochastic setting. Our work suggests that Lion has the ability to effectively balance fast and slow averaging, leading to stable and rapid convergence.