#cost-governance

Test cheaper models with open-weight benchmark harness

A benchmark harness evaluates open-weight, closed, and local models on real product tasks, preventing hidden costs when traffic is switched, by computing a cost-per-success metric [Dev.to].