Partial plant knowledge may be used to develop model-based extremum seekers; however, existing results rely on a type of time-scale separation which leads to slow optimization relative to the plant dynamics. In this work, a fast model-based extremum seeking scheme is proposed for a Hammerstein plant, and semi-global stability results are provided. The structure of a Hammerstein plant is used to advantage in designing filters that enable the extremum seeker to act on a faster time-scale than the plant dynamics. This leads to fast convergence while maintaining semi-global stability.