← Back to the lab
Organizing context beats switching models
I tested how to choose what goes into the prompt, in what order and where each piece of information is stored, before thinking about switching models.
76%of input tokens come from cache
What the measurement showed
- 76% of chat input tokens come from cache; 40–58% in scheduled routines.
- The cost panel overestimated input by roughly 2.5–3× because it ignored the cache.
From the OrganIA lab (agent-builder).
● Lab Log
One note from the lab every two weeks.
Experiments, real numbers and what went wrong. No spam, leave anytime.