Small models, big latency budgets

Not every problem needs a frontier model. Wren compares small, locally run models against large hosted ones on three real tasks, measuring latency, cost and the failure modes nobody puts in a benchmark.

Learn a decision tree for choosing model size and where to cache.