Watch reasoning tokens though. We tried a small reasoning model that burned ~2800 thinking tokens per call, 3x the cost of a cheaper non-reasoning one despite a

Watch reasoning tokens though. We tried a small reasoning model that burned ~2800 thinking tokens per call, 3x the cost of a cheaper non-reasoning one despite a better price sheet.