For ~equivalent tasks/results, or because we’re expecting more or better from tasks?
replyThe real measure should be cost per ~equivalent task result, not cost per token nor tokens per task.
For better performance of ~equivalent tasks. That's what all the harness tooling people are using does: (often) increasing output quality by significantly increasing token counts.
reply