Hint: highest possible is not an acceptable answer
Another way to ask about this is: How would you decide what an appropriate predictive accuracy is for your model? 

The context of the problem is very important in deciding whether the model accuracy is good enough or not. For example, in case of binary classification if we are more concerned on True positive cases and we are getting good TPR rate then we are good to go but if we are building a model to detect fraud detection then the accuracy of 92% or greater is not what we need. We are concerned of those cases when fraud actually happened(True Negative) and in these scenarios, we might have low data points for cases where fraud happened and we can use (undersampling or oversampling or SMOTE) for making a more accurate model to detect fraud.

One cannot answer this question based on simply this information.I would ask follow up questions like – What is the distribution of the classes. if they are fairly similarly distributed (i.e. each class has more or less the same number of observations) then accuracy can be considered. However, if not (i.e. one class has a lot many samples than the other class) accuracy is misleading and not a good measure. What dataset was this accuracy calculated on ? If it is the training accuracy, there might be chances that the model is overfit and won\\\’t perform as good on the test set. Were there any other models that were fit? is 92% highest test accuracy?