TokenCost.dev | AI API Cost Calculator

Workload assumptions

Define one representative request.

Beginner mode: use the presets and plain-language examples. Advanced limits and pricing controls are hidden, but the estimate still uses safe defaults.

Input tokens per request

500

Includes instructions, previous messages and any document text.

100: short question2,000: short article10,000+: long document

Output tokens per request

300

The average length of the AI-generated answer.

50: 1–2 sentences250: short explanation1,000+: detailed answer

Request volume

Volume period

Daily estimates use a transparent 30-day month.

Peak requests per minute

Used only for free-tier feasibility.

Display currency

No live exchange-rate feed is used.

Optimisation assumptions

Discounts apply only where the model data says they are supported.

Model prompt caching

Asynchronous batch processing

Applied only to models with explicit batch rates in this beta dataset.

Current scenario

Monthly requests

10,000

Input tokens

5.00M

Output tokens

3.00M

30-day average/day

333

Estimated monthly comparison

Free-tier rows are ranked only when the scenario fits their structured request and token limits.

Detailed results

Effective rates reflect the currently selected assumptions.

Model	Tier	Base input	Effective input	Base output	Effective output	Free-tier feasibility	Cost/request	Monthly estimate

Model catalogue

A deliberately small beta dataset. Values should be maintained from provider sources.

Dynamic configuration export

Choose a model and generate a safe .env.example. Placeholder keys are never real credentials.

Primary model

Methodology and limitations

What the beta includes—and deliberately does not claim.

Token cost: monthly tokens in millions × the model's effective input/output rates.
Prompt caching: base rate remains for non-cacheable input and cache misses; cached-input rate applies only to the cacheable share that hits.
Batch: explicit batch input/output rates replace normal rates only for supported models.
Free tiers: feasibility is evaluated using structured requests/day, requests/minute and tokens/minute limits where present. A free tier that does not fit is not treated as a valid £0/$0 recommendation.
Not included: taxes, retries, tool calls, image/audio charges, embeddings, vector databases, provider credits, regional pricing, long-context price bands and self-hosted GPU TCO.
Currency: source list prices are stored in USD. GBP display uses a fixed illustrative conversion and is not a live FX quote.

Beginner glossary

Plain-language explanations of the terms used in this calculator.

Token: A small piece of text used by an AI model. Tokens are not exactly the same as words; 100 English words may be roughly 130–150 tokens.
Prompt: Everything sent to the model: instructions, the user's question, prior chat messages and any included document content.
Input tokens: Tokens sent to the model. Providers normally charge separately for input and output.
Output tokens: Tokens generated by the model in its answer. Longer answers usually cost more.
API request: One call from an application to an AI provider, such as one chat message or one document summary.
Context window: The maximum combined amount of text a model can process in one request, including input and generated output.
Prompt caching: A lower price for eligible text that is repeated across requests and successfully reused by the provider.
Cache hit: A request where previously cached text is found and reused. Cache misses are charged at the normal input rate.
Batch processing: Submitting non-urgent requests for later completion, often at a discount. It is unsuitable for immediate conversations.
Rate limit: A provider cap on requests or tokens over a period. RPM means requests per minute, RPD means requests per day and TPM means tokens per minute.
Base rate: The provider's standard price before any eligible caching or batch discount.
Effective rate: The estimated rate used by this calculator after applying the selected, supported optimisation assumptions.
Free-tier feasibility: Whether your projected workload appears to fit the provider's stated free request and token limits.
Monthly estimate: An approximation based on your assumptions. It may exclude taxes, retries, images, audio, tools and other charges.

Compare AI API costs in seconds.