Optimization by fmincon with central finite differences gradient

I am using (MATLAB version R2015b,fmincon function, sequential quadratic programming algorithm) to identify a nonlinear state space model using MLE and Kalman filter. I used an UKF to calculate The log-likelihood expression: http://stats.stackexchange.com/questions/249626/negative-loglikelihood-kalman-filter.

fmincon is a gradient-based optimization method, I used a supplied gradient of the negative log-likelihood function to find the optimal point. Now I want to compare my results and the results found by using gradient estimated by central finite differences, To know which method is more time-consuming than the other.

I performed multiple optimizations runs, the optimization with my supplied gradient takes approximately the same time at each run, But with central finite differences gradient, I Get a different time at each run.