Most AI API pricing pages quote a rate per million tokens, which doesn't answer the question that actually matters when you're budgeting: what does one call cost, and what will this cost at the volume you're planning to run? This calculator converts a model's published per-token rate into a per-request and projected monthly cost, using your own token counts.
How it works
Every major provider bills input tokens (what you send: prompt, system message, and context) and output tokens (what the model generates) at different rates: output is almost always priced several times higher than input.
The calculator multiplies your input token count by the model's input rate, and your output token count by its output rate, both divided by 1,000,000 since rates are published per million tokens.
Multiplying the per-request cost by your expected requests per month gives a monthly projection, useful for comparing models before committing to one in production, or for catching a cost blind spot before it shows up on an invoice.
A worked example
A request with 5,000 input tokens and 1,000 output tokens on Claude Sonnet 5 ($2/M input, $10/M output) costs about $0.02 ($0.01 input + $0.01 output) per call. At 10,000 requests a month, that's roughly $200/month, a number worth knowing before, not after, you ship a feature that calls the API on every page load.
Questions people ask
Why is output so much more expensive than input across every provider?
Generating each output token requires a full forward pass through the model, run sequentially, one token at a time. Input tokens, by contrast, can be processed in parallel in a single pass. That computational difference is why output pricing runs roughly 4–6x input pricing across most current models.
How current is the pricing used here?
The rates in this calculator were checked against each provider's official pricing documentation and cross-referenced with independent trackers in September 2026. AI API pricing changes frequently, sometimes with only weeks of notice, so treat this as a close estimate and confirm the exact current rate on the provider's own pricing page before finalizing a budget.
Does this include prompt caching or batch API discounts?
No. This calculator uses each model's standard, non-cached, non-batch rate. Prompt caching (for repeated system prompts or context) and batch processing (for non-real-time workloads) can both cut costs significantly, commonly by 50–90% on the portions they apply to: worth checking if your workload fits either pattern.
Why isn't every current model listed?
This calculator lists a representative set of current models from OpenAI, Anthropic, and Google rather than every historical or specialized variant. If you need a model not listed here, you can calculate it manually: (input tokens ÷ 1,000,000 × input rate) + (output tokens ÷ 1,000,000 × output rate).