My bet on the fastest way to do this is to use CREATE TABLE AS SELECT with its nologging and parallel features to create a new table containing the 32 million rows that you still want, and then drop the old table, rename the new one and build the required indexes and so on. You could even make it a partitioned table in the process (using the EXCHANGE PARTITION syntax) so that you don't end up with the same problem again. I did some tests a while back that showed CTAS to be about 30% faster than an equivalent INSERT /*+ APPEND */ which was the next best option.

What is the best approach to delete 14 millions rows from a table with 46 millions rows? The table is not partitioned.