AI API Cost Calculator

Estimate token costs across popular AI models in under a minute. Start with a workload preset or enter your own usage.

Pricing checked June 2026

Start with a workload

Choose a realistic starting point, then fine-tune the inputs below.

How many requests you send to the model each month. One call is a single request and response — one prompt in, one completion out. Your total token volume, and cost, scale directly with this.
The size of each request you send: your prompt, system instructions, conversation history, and any retrieved or pasted context. Roughly 1 token ≈ 4 characters (about ¾ of a word). Billed at the model's input rate.
The length of the model's reply per call — everything it generates back, including any reasoning. Output tokens are usually several times more expensive than input tokens, so this strongly affects the total.
The share of your input tokens served from the provider's prompt cache. Repeated context — system prompts, long documents, chat history — can be cached and re-read at a fraction of the normal input price (here: Claude ~10%, OpenAI ~50%, Gemini ~25%). A higher % means lower input cost. 0%
Batch APIs handle requests asynchronously — you submit a batch and collect the results later (often minutes, up to 24h) instead of getting an instant reply. In return, providers charge about 50% less per token. Ideal for large, non-urgent jobs like bulk summarization, classification, or evals — not for chat or anything a user is waiting on.
Not sure about token counts? Estimate from text
0 tokens · 0 characters

Rough estimate (~4 characters per token). Actual tokenization varies by model and language.

Calls / month
1K
Input tokens / mo
1.0M
Output tokens / mo
500K

Plan from a monthly budget

See how many calls each model can handle with your budget and the token sizes above.

Rates are USD per 1M tokens, as of June 2026. Provider prices change often — always verify on each provider's pricing page: Anthropic, OpenAI, Google Gemini. Estimates cover API token usage only — not hosting, storage, or other infrastructure.