AI API cost calculator

Enter how much text you send and get back, and see what a month costs on each model. Prices come from each company’s own pricing page.

Prices checked October 13, 2026

Start from an example:

Count in

TokensWords

A token is a piece of a word. For English, 100 tokens is about 60 to 80 words (Google).

Requests a month

In per request

Out per request

In: your prompt, instructions and any pasted text. Out: the answer, including any thinking the model does before it answers.

Input read from cache: 50%

The part of each prompt that repeats, such as fixed instructions, billed at the lower cached rate.

Prices

TodayAnnounced rises

Google lists higher prices for Gemini 4 Argon from 1 December 2026 and for Gemini 3.8 and 3.6 Flash from 1 January 2027.

Send as a batchFor requests that can wait, at the company’s batch price where it has one.DeepSeek off-peak hoursHalf price outside 01:00–04:00 and 06:00–10:00 UTC on weekdays.

Monthly cost

Models

Turn on JavaScript to see the costs. The price tables below work without it.

Prices by company

Claude API pricingAnthropicfrom $0.10 in · $0.50 outClaude Haiku 5.5, per 1M tokens GPT API pricingOpenAIfrom $0.10 in · $0.50 outGPT-6 Luna, per 1M tokens Gemini API pricingGooglefrom $0.30 in · $2.50 outGemini 3.5 Flash-Lite, per 1M tokens Grok API pricingxAIfrom $2.00 in · $6.00 outGrok 4.7, per 1M tokens DeepSeek API pricingDeepSeekfrom $0.30 in · $1.20 outDeepSeek Flash, per 1M tokens

Comparisons

Claude Opus vs SonnetOpus 5.5 costs twice as much as Sonnet 5.5 GPT-6.1 Sol vs GPT-6 AstraAstra costs five times as much Claude Haiku 5.5 vs GPT-6 LunaSame list price, different long-prompt rules Nano Banana pricingGemini image models, per image

How the numbers work

AI companies charge for the text a model reads (input) and the text it writes (output), counted in tokens and priced per million. Output costs three to eight times as much as input (five times on most current models), so long answers move the bill more than long questions.

The calculator works out, for each model: input not read from cache × input price, plus cached input × cached price, plus output × output price, divided by one million. With batch on, it uses each company’s batch price instead.

Some models charge more once a single prompt gets long:

Claude Haiku 5.5: over 100,000 tokens
GPT-6, GPT-5.6, GPT-5.5 and GPT-5.4: over 272,000 tokens
Grok 4.7: 200,000 tokens or more
Gemini 3.1 Pro and 2.5 Pro: over 200,000 tokens

The calculator switches to the higher price when your input per request passes that point. Try the “Long documents” example to see it.

Not included: cache-write charges, Gemini’s hourly cache storage fee, web search and other tool fees, image and audio tokens, taxes, and regional or data-residency surcharges (about 10% at Anthropic and OpenAI). Your real bill can be higher.

Questions

What is a token?

A piece of a word. Google’s rule of thumb for its Gemini models is about 4 characters per token, or 60 to 80 English words per 100 tokens. Each company splits text its own way, so the same text can be a different number of tokens on different models. The token counter estimates it for your own text.

Which model is cheapest?

It depends on the job. In the support-chat example, Claude Haiku 5.5 and GPT-6 Luna cost the same at list prices, because both charge $0.10 per million input tokens and $0.50 per million output tokens. With 150,000-token documents, Haiku 5.5 moves to its long-prompt price and GPT-6 Luna does not. Change the numbers to match your own use.

Should I count in tokens or words?

Tokens, if your logs or the company’s dashboard give you real counts. Words, for a first estimate. In word mode the calculator adds 30% for Claude 4.7 and later models, because Anthropic says their newer tokenizer produces about 30% more tokens for the same text. You can switch that off.

How often are prices checked?

Every week, against each company’s pricing page. Changes and announced future prices go in the price change log with their dates.