Cross validation score/Training score/Test score : what should i considered to say whether a model is a well generalised model?

Question

I am new to Machine learning domain and I want to clear my doubt. My model is a multi class classification model based on smiles notation dataset. And my dataset is less than 1000 rows and also it is an imbalance dataset. Suppose i am getting high training score of 97% and cross validation score of 80% and test score of 82%. To say that my model is a well generalisation model, what should i take, whether i compare training or test score to say that my model is a good or well generalised model or compare test score or cross validation? I know that the training score reflects how well your model is fitting the training data, and but I read somewhere that there should be less difference between training and test score for a well generalized model.