ThePrimeagen burns 56% of a $500 plan in an hour
ThePrimeagen bought Claude's $500 weekly plan, prompted for about 60 minutes, and says it ate 56% of his limits. If you are weighing a subscription against the API, that is the number to price against.
Receipts
ThePrimeagen says an hour of prompting ate 56% of a $500 plan
ThePrimeagen bought the $500 weekly plan and says roughly 60 minutes of prompting consumed 56% of his weekly limits. In a follow-up he added that he now sees why people hate Astra, which suggests the burn came from a model he did not pick deliberately. He did not say what he was building.
I bought the $500 plan today, prompted for ~60 minutes
Cloudflare ships two decision models, Steinberger notes how fast this spread
Cloudflare released two of its own trained decision models, Clef and Clef-flash, which Peter Steinberger flagged with a one line reaction. He says he has never seen an idea spread so fast. He did not report testing them.
Never seen an idea spreading so fast.
How they actually work
Pocock's prompt of the day: read my last 10 agent sessions
Matt Pocock posted a /retro prompt that asks the agent to read his last 10 coding sessions and find where agents took too long to locate information or relied on out of date docs. The output is a list of changes to make the repo easier to navigate. He calls navigability an underrated way to save tokens.
Improving navigability is such an underrated way to save tokens.