I compared Jev, which TypeSafe AI describes as calibrated, with three small LLMs. It was cheaper and competitive on accuracy, but no better calibrated.