Ask HN: How do you know a prompt is "complex" for an AI model?

By faryadz · 2026-09-07 · 1 points · 0 comments

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.

Open the full discussion on BetterNews