Home › Calculators › Cost Comparison
Does running it yourself pay off against billing per request?
Running it yourself pays off from a regular usage volume upwards, because costs are then decoupled from the number of requests. Exactly where the break-even point lies depends on model size, utilisation and term. Enter your own figures below, all assumptions are disclosed.
Your figures
Result
€4,000 / mo
Cost of running it yourself (hardware + ops)
€675 / mo
Cost of commercial pay-per-request API
1,481,481 req/mo
Break-even threshold at current token length
Our own cost model, not a study figure. All assumptions in the section below.
The assumptions, disclosed
| Item | Assumption | Source |
|---|---|---|
| Hardware | purchase or lease, depreciated over the chosen term (36 mo) | public hardware price lists |
| Power and facilities | consumption at assumed utilisation, location Germany | our own measurement |
| Staff & Operations | operating share per service tier, not a full-time role | our own calculation |
| Comparison price | list price per token from common providers, as of retrieval date | provider websites |
| Not included | one-off implementation, connection to your systems, training | see pricing |
Important: inference costs have fallen sharply over the past two years. Any static calculation goes stale, which is why we state the retrieval date and update quarterly.
Pricing and packages → · Private AI infrastructure → · To the comparison article, if you would rather read →