Evaluating Regression models from a same data set

let's assume I created Three Regression models out of same data set.

first model: This is created using the whole dataset(of course diving it into test and train) and R2 = .98 and RMSE = 3.2.

Second model: This is created with a part of the data set( divided based of a category) and used to predit the data of this category only. r2 and RMSE is .96 and 1.5 respectively.

Third model: this is also created similarly with the remaining category in the dataset and will be used to predict that category only. R2 and RMSE are .97 and 1.2 respectively.

So in this case, which approach should I prefer...creating a wholesome model for the entire dataset or do individual models for two categories and use that respectively for predictions? And what will be the combined R2 and Rmse if I choose to go with doing separate models?

Thanks for responding.