We are not able to run your model without the various parameter assignments (do you use a .m file to initialize variable values?) and XML files (though we can assume you use the default ones from Wolfspeed). We would need this information to fully determine the validity of your claims about the high efficiency.
Having said that, we recommend you use the Switch Loss Calculator block in the PLECS library and provide the switching frequency as the averaging period in your case, rather than doing your own filtering for loss averaging.
Further, because of the way that PLECS works where the ideal switches are simulated in parallel to the thermal loss calculations, you are suggested to use (1-Psw/Pin)*100% for the efficiency formula. We have a demo model "Buck converter with thermal model" that is a great template for loss calculation, by the way.