247 AI agent sessions cost about $178 for the week
A firsthand audit counted more than 247 Claude Code and Codex sessions across five computers in one week. The setup used two Anthropic accounts and ChatGPT Pro, processing 19.59 billion and generating 61.8 million . At standard API prices, that activity was estimated at about $19,700.
The actual cost came from $600 in plus $40 of extra for company work, giving an estimated weekly cost of about $178. The API-price estimate was therefore about 111 times the amount actually spent that week. About 94% of all token activity was cache reads, so most of the 19.59 billion operations did not involve generating fresh material.
Saturday alone reached 13.83 billion , or roughly $13,527 at API prices. The agents handled coding, audits, testing, research, , genealogy, workflow design, and fictional-language development. The largest Claude session ran for more than 22 hours and added nearly 30,000 lines; another request for five subagents unexpectedly launched 135, yet the resulting work reportedly stayed largely coherent.
Key points
- More than 247 agent sessions ran across five computers in one week.
- The agents generated 61.8 million , while cache reads accounted for about 94% of total activity.
- About $178 in estimated weekly spending handled usage valued at roughly $19,700 under standard API pricing.
- The longest Claude session ran for over 22 hours and added nearly 30,000 lines.
- A request for five subagents launched 135, showing that automatic expansion needs monitoring.