Performance of Savepoint vs Commits when looping through large datasets

 We are running Oracle 11g (11.2.0.1.0) database and I have a PL/SQL process that loops through records in an interface table, calls a API for each record, depending on the success of the API, additional processing occurs.

I am wondering what the recommendations from others who have done this sort of thing on using commit and rollback if there is a failure from API vs Savepoint rollback?

Simple code example

Option with Commit/Rollback

Loop through table

   Call API (l_item, l_status, l_message);

   if l_status != 'S' then

     rollback;

  end if;

  update table with status;

  commit;

end loop;