The first that comes to my mind is this simple insert test run on an Amazon RDS. But I think that those inserts are performed with the extended insert syntax with several rows per statement.

Which compared to my 2.5 year old laptop which manages to average about 5,000 inserts per second with a single thread single row per statement insert (as long as the disk is not too heavily used ), with innodb_flush_log_at_trx_commit=2 to avoid too much IO congestion.

So there you have some ballpark figures that basically doesn't say anything since the hardware can be quite different and you basically never run a server without reading and updating the data simultaneously.