Phelps McManus
<Silver Donator>
- 265
- 257
Think about a place like nvidia where jensen said if your senior devs aren't using 250k tokens they shouldn't be employed. 250k tokens is more than we have allocated for our entire couple hundred developers. We get between 100-700 tokens a month.
Something is off on your scale or maybe you are taking about context sizes? I burn through a million tokens ($3 of Claude sonnet) in around 1-2 hours, and I am fairly diligent about clearing my context between prompts. If Claude says it is compacting, that means you hit the 50k context limit, which is eating up over 50k tokens on that one prompt (50k in plus whatever compacted version comes out). They have 1mil context models and I cannot imagine the burn rate for people who use them.