AI Token Cost Calculator
Adjust the inputs below. Results update as you type.
How it works
AI APIs charge per token — roughly 750 tokens per 1,000 words of English. Input tokens (your prompt) and output tokens (the model's response) are priced separately, with output typically 3–5× more expensive. Costs scale linearly with usage. Techniques like prompt caching, batching, and choosing smaller models for simple tasks can reduce costs significantly.
Input guidance
- Confirm date/time/unit settings before comparing outputs.
- If a task has optional fields, run both with and without them to understand impact.
- Save scenario variants when making practical decisions.
The formula
Cost per request ≈ (input tokens × input price per 1M ÷ 1,000,000) + (output tokens × output price per 1M ÷ 1,000,000). Monthly cost = cost per request × requests per day × 30, with optional cache-hit discounts on repeated input.
Worked example
A call with 2,000 input and 500 output tokens at $3/$15 per million tokens costs about $0.0135. At 1,000 requests/day that is roughly $405/month before caching.
More examples to test
- Quick-pass example: run default assumptions for a first estimate.
- Refined example: update one assumption at a time to isolate impact.
How to interpret results
Use outputs as operational estimates and pair them with local constraints, business rules, or provider requirements.
When this can be inaccurate
Utilities can miss local policy details, special-case rules, and environment-specific constraints.
Change history
- July 2026: Published AI token cost path with model-rate and usage caveats.
Site-wide corrections also appear on the corrections log.
Find productivity tools for everyday workflows
General recommendation, not professional advice.
Frequently asked questions
Why are output tokens more expensive?+
Providers usually price generated tokens higher than prompt tokens because generation is more compute-heavy. Shorter replies often cut spend faster than trimming the prompt alone.
How does prompt caching help?+
When a large system prompt repeats, cached input is billed at a lower rate. A high cache-hit percentage can shrink monthly API cost without changing the model.
Are list prices stable?+
No. Model SKUs and per-token rates change. Treat this estimate as a planning tool and confirm current provider pricing before locking a production budget.
Why are outputs slightly different from another tool?+
Different tools can use different rounding rules, default assumptions, and treatment of edge cases.
How can I improve estimate reliability?+
Use verified inputs from real records and re-run calculations after major assumption changes.