Technology
Estimate monthly inference cost from token counts, call volume, and prices you enter per million tokens. Prices are your inputs, not live provider prices.
Results update as you type. No calculate button. All calculations run locally in your browser.
The token prices are example inputs, not live prices for any provider or model. Replace them with the prices that apply to your own account before using the result.
—
—
| Item | Value |
|---|
Monthly tokens are calls times tokens per call. Prices are applied per million tokens, as entered.
Bars are scaled to the larger cost component and show whether input or output tokens dominate the bill.
| Scenario | Value |
|---|
Each row reuses your tokens per call and prices and varies only the call count, so cost scales linearly with volume.
Monthly cost = calls × (input tokens × input price + output tokens × output price) ÷ 1,000,000, with prices in USD per million tokens. Cost per call = monthly cost ÷ calls.
Input 2,500 tokens, output 800 tokens, 100,000 calls, input price $5 and output price $20 per million tokens. Input cost = 100,000 × 2,500 × 5 ÷ 1,000,000 = $1,250. Output cost = 100,000 × 800 × 20 ÷ 1,000,000 = $1,600. Total = $2,850 per month, or $0.0285 per call. The prices in this example are illustrative only — enter your own plan's prices for a usable figure.
Prices are entered per million tokens and are not live provider prices; the defaults are illustrative, not a quote for any model. Cached tokens, batch discounts, retries, failed calls, rate-limit retries, and infrastructure or staffing costs are excluded. Token counts are your averages: real calls vary, and a few long outputs can move the average materially. Taxes and currency conversion are not included.
Each call's cost is input tokens times the input price plus output tokens times the output price, divided by 1,000,000; that per-call cost times the monthly call count is the monthly total.
No. They are example inputs. Provider prices change and differ by model and tier, so enter the prices from your own plan or invoice before using the result.
Output prices per token are usually higher than input prices, and long outputs add up. The breakdown table shows the split so you can see which side to optimise.
Cached or discounted tokens, batch pricing, retries, infrastructure, and staff time. If those are material for you, add them separately rather than treating this total as a full cost of ownership.
Measure a sample of real calls: record input and output token counts, average them, and re-check after prompts change. Averages from a handful of short test calls usually understate production calls.
No. It is an estimate from your averages and prices. Compare it with an actual invoice for one month, then adjust the token averages or prices until the model tracks reality.
Results are estimates only and do not constitute professional advice.