What will your AI feature actually cost?
Describe the workload once — see the monthly bill across every major model, with prompt caching factored in. No signup, no email, just the math.
Estimated monthly cost by model
Hover a bar for the input / output split. Cached input billed at ~10% of list price.
List prices last verified August 2026. Batch APIs (often −50%), committed-use discounts and regional pricing can change the picture — treat these as directional estimates.
How the math works
Monthly cost = 30 days × requests/day × (input tokens × input price + output tokens × output price), using each provider's standard list price per million tokens. The prompt-cache slider models the share of your input that's identical across requests — system prompts, knowledge bases, few-shot examples — which providers re-serve at roughly 10% of the list input price. Two things this deliberately ignores: batch APIs (often another 50% off for non-realtime work) and committed-use discounts — so treat these numbers as the honest ceiling, not the floor.
The spread between cheapest and priciest is usually the most useful number on this page: it's the budget you free up by routing easy traffic to a small model and saving the frontier model for requests that need it.
Not sure which models to shortlist? Start with our opinionated picks per job.
Which model for which job? →This math moves with caching, batching, routing and model choice. Ahitrisan Intel designs AI systems that hit the quality bar at a fraction of the naive cost — this calculator is the first five minutes of that conversation.
Get a cost review →