ThePrimeagen burns 56% of a $500 plan in an hour
One hour of prompting, more than half a week of allowance gone, and a follow-up pointing at Astra.
ThePrimeagen bought a $500 plan, prompted for about 60 minutes, and says it ate 56% of his weekly limits. That is the whole report. He did not say what he was building, and he did not name the provider.
I bought the $500 plan today, prompted for ~60 minutes
The follow-up is the interesting part. Minutes later he posted that he now sees why people hate Astra, which reads like the burn came from a model he was not deliberately steering toward. He did not spell out the connection, so treat the link between the two posts as his implication rather than a stated cause.
What it is and is not
This is one person, one hour, one unspecified workload. It is not a measurement of what the top tier buys in general, and nobody has published a controlled comparison. What makes it worth passing along is that hard numbers on premium plan consumption are rare, and most of the public argument about limits runs on vibes and screenshots of rate limit errors.
It also lands in a week where plan economics keep coming up from both directions. Theo previously worked out that a $200 Claude subscription had bought him roughly $9,000 of Opus usage at API prices, and separately OpenAI halved what $200 of Pro buys. Those are different vendors and different plans from the one ThePrimeagen bought, so they are context, not corroboration.
For working engineers the practical read is narrow. Spend per hour on agentic work is dominated by which model actually ends up doing the turns, and a heavy hour on an expensive model is not comparable to a heavy hour on a cheap one. ThePrimeagen's two posts together suggest he found that out by watching the meter rather than by reading the pricing page, which is how most people find out.
also... i see why people hate astra...