Cost becomes the limit before performance does
AIMade by AISummarised by AI (Claude) from one piece of material it read (a talk or an explainer).How it's made
In a comparison that ran the same task with only the depth of thinking changed across five levels, the converted cost rose from $1.45 to $30.35 on one task and from $3.95 to $101.06 on the other, and on both tasks the biggest jump came in the step up to the highest level. The highest level ranked first on quality, but the gain in quality was small next to the rise in price, so the conclusion was to stay at a low level when only the function is needed and that the stretch with the best value for money is the upper-middle level. The write-up itself noted the limit that this is its own measurement, taken only once each. A write-up on cutting cost holds that which model the work goes to, how deeply it is made to think and how many files it is made to read matter more than the length of the prompt. There is even an account of someone building a device that shows the money being spent on the spot, like a meter, so making cost visible is itself becoming a feature of the tooling.
Key points
- A self-reported measurement in which raising the depth level made the converted cost jump more than 20-fold
- The stretch with the best value for money is not the highest level
- Which model, how deep, and how many files are read decide cost more than prompt length
- A gauge that shows the money being spent on the spot, added by hand
This is a note: a summary an AI (Claude) wrote of one piece of material it read (a talk or an explainer). It has no source link; events checked against a source are kept as records.