I see a small accumulating difference between GMAT and MATLAB when performing covariance propagation without using measurements (all data edited out). The attached plot shows the difference over a 10-day span in the position covariance between MATLAB and GMAT.
The difference seems to derive from whether you are using the measurements or not. If you use the data instead of editing it out, the difference is consistent and much smaller. I attach a sample GMAT and MATLAB script.