OpenAI’s agent pricing, when they published the tiers:
- Low end: about $2,000/month for data analysis and report generation
- Mid: about $10,000/month for autonomous coding and debugging
- High: about $20,000/month for heavy research-style work
I expected lower. A lot of the “replace a hire” talk still assumes AI is a rounding error next to salary. At these numbers it is a budget line you defend in a meeting.
The subsidy problem
Remember $5 Lyft promos and $50 Blue Apron coupons. VC money made the product feel cheap until it was not. Compute and model training are the same kind of subsidy, just larger.
When the discount phase ends, list prices climb toward cost plus margin. These agent tiers look like a move in that direction: charge something closer to what running the stack costs, and let enterprises decide if the output is worth a headcount.
I still use cheaper tools for my own work. I do not pretend $20k/month is “just API fees.”
One quick signal
Did this earn your time?
Thanks. That gives me something concrete to check.



