What will your AI feature cost to run each month?
Inference cost is driven by four things: how many requests you serve, how much context each one carries, which model answers it, and how much of that is cacheable. Set those four and this calculator gives a monthly figure in INR and USD, plus the cost per request, using published provider pricing.
Most AI budgets get the build right and the running cost wrong. This is the model we use in Sprint Zero, with the same assumptions written down.
Show in
Assumptions
- Prices are per million tokens and change often; check the provider's page before committing a budget.
- A token is roughly four characters of English. Indian languages in Latin script run 15–30% longer.
- Caching applies to repeated prompt prefixes — system prompts, retrieved passages, few-shot examples. A well-built RAG system commonly caches 40–70%.
- This covers model inference only. Speech, telephony, vector storage and hosting are separate lines.
Want this checked against your real numbers?
Send your details and we will model it properly: your volumes, your languages, your deployment, and the parts this calculator deliberately leaves out.