The client should always be prepared to handle an error returned from the API that indicates that the access_token validation failed. Depending on the implementation the access token may have been revoked or declared invalid otherwise.
The client may then use a refresh_token to get a new access token and try again. So you can choose to implement 1. but it does not free you from implementing 2. as well, so you may choose to stick to only implementing 2 and minimize the amount of code required.
Of course if you want to prevent errors from happening as much as possible you could implement 1. to optimize the number of calls and reduce the number of errors in the whole process.