Number of tokens is still a useful metric, as their endpoints have Tokens Per Minute quotas. Decreasing number of tokens used means increasing throughput, up until the Request Per Minute quota.