feat: benchmark multiple models and validate improved results

7b233d3