3 ms·
Even more than that in practice once you factor in prompt caching
by weird-eye-issue 7mo ago
Even more than that in practice once you factor in prompt caching
- kinlan 7mo agoI think we still skew back to an insanely high input token ratio when you consider agentic loops. For example, when I see the tools I use do a web fetch or a search or other tool use, it's an incredibly high number of new input tokens.