benchmark

24b35547 - run more iterations for short models