问HN:你如何判断一个提示对AI模型来说是“复杂”的?

1作者: faryadz16 天前原帖
我正在努力在我的应用程序中平衡API成本和延迟。目前,我默认使用前沿模型(Claude Sonnet),因为它们可靠,但我知道我在一些较小的任务上支付了过高的费用,这些任务可能由较小的模型(如Haiku或轻量级开源模型)来处理。
查看原文
I’m trying to balance API costs and latency in my applications. Right now, I default to frontier models (Claude Sonnet) because they are reliable, but I know I’m overpaying for tasks that smaller models (like Haiku, or lite open source models) could likely handle.