This paper presents a unified analysis of three well known extra gradient methods that require only one gradient query for each step, in deterministic and stochastic settings obtaining the tight O(1/t) convergence rate.