Should I use a different ML model for feature sets with missing values while drawing inferences?

I have trained a model with 5 features. While training, I had no missing values, and the models were perfectly trained. However, when I’m using this model to draw inferences, sometimes I’m missing one feature value. How can I address this – should I use a different model with that only uses 4 features, or should I keep using the model with 5 features and draw the inference using the available 4 features?