AI application profitability

Attempts/user = requests/user × (1 + extra attempts%/100). Weighted input rate = standard rate × (1 − cached share%/100) + cached rate × cached share%/100. Model cost/attempt = (input tokens × weighted input rate + output tokens × output rate) ÷ 1,000,000. Contribution/user = price × (1 − payment fee%/100) − attempts × (model cost/attempt + other cost/attempt) − other cost/user. Monthly profit = paying users × contribution − fixed costs. Break-even users = ceil(fixed costs ÷ contribution) when contribution is positive. Use monthly figures for paying users, subscription price, baseline requests and fixed costs. Enter average input and output tokens per attempt, including context you actually send, and use provider rates per million tokens in the same currency as your revenue. The cached percentage is the share of input tokens billed at the cached rate. Extra attempts add retry or repeated-call usage above baseline requests; 10% means 110 attempts for every 100 baseline requests. Every attempt incurs the entered model and other per-attempt costs. The result combines those costs, per-user costs, payment fees and fixed overheads to estimate operating profit and the user count needed to cover fixed costs.

With paying users per month 1,000, monthly price per paying user 15, baseline requests per user per month 200, average input tokens per attempt 1,000, average output tokens per attempt 500, input price per million tokens 3, cached input price per million tokens 0.3, output price per million tokens 15, input tokens billed at cached rate (%) 50, extra attempts per 100 baseline requests 10, other cost per attempt 0.001, other monthly cost per paying user 1, fixed monthly operating costs 500, payment fee on subscription revenue (%) 3, the monthly operating profit is $10,817.00. Monthly revenue: $15,000.00; Monthly model cost: $2,013.00; Total monthly operating costs: $4,183.00; Contribution per paying user: $11.32; Break-even users: 45; Operating margin: 72.11%.

This is an estimate based on the inputs and assumptions shown on this page, not professional advice. Check anything with financial, health or safety consequences independently.

Yes. Calculations run in your browser and your numbers are not sent to a server. A share link contains the inputs you choose to share, so check it before sending it to anyone.

Yes. Pick from 33 currencies and the amounts are labelled to match. The selector does not convert between currencies, so enter prices in your own currency.

Calculate revenue from paying users, then subtract model usage, other per-attempt and per-user costs, payment fees and fixed monthly costs. This calculator assumes a constant average usage level per user, so compare heavier-usage scenarios before relying on a single estimate.

It is the share of input tokens billed at your entered cached-input rate. If half your input tokens use that rate, enter 50. It is not the percentage of requests with any cache hit. Output tokens keep their output rate, and separate cache-write or storage charges are not modelled automatically.

The extra-attempts field adds attempts per 100 baseline requests. Entering 10 adds 10 extra attempts, giving 110 total. Every attempt incurs the same average tokens and other per-attempt cost. The field is not a failure probability and does not simulate repeated retries until success.

It means one paying user's net revenue does not exceed their variable costs under your inputs. Adding more identical users cannot generate a positive contribution to fixed costs. Try changing price, usage or costs, then inspect contribution per user again.

No. All default rates are illustrative. Copy the applicable rates for your model, service tier and billing currency from your provider's current pricing documentation. Include all billed input and output tokens, and account separately for charges the model does not cover.

Only costs you enter are included. The user count represents paying users. Allocate free-tier usage, acquisition spending and other costs to suitable per-user or fixed-cost fields if you want a broader estimate, taking care not to count a cost twice.