TapCub is live — analytics, insights and live chat, free on one platform

Free tool · AI LLM cost calculator for OpenAI, Claude, Gemini and more

Enter how many requests you expect and how long prompts and answers are. The calculator prices them with each provider's published per-token rates and ranks the models by monthly cost, in US dollars or Chinese yuan.

  • List prices with source links
  • USD and CNY side by side
  • Runs in your browser

Estimate your monthly API bill

Change any number or tick another model and the table re-sorts. Prices are per million tokens, standard (non-batch) calls.

Calculator

Usage

API calls, not visitors: one chat answer is usually one call.

System prompt, history, retrieved text and the question.

The answer, including reasoning tokens.

Share of input read from the provider's prompt cache.

Count tokens from a sample prompt and answer

Paste a typical prompt and answer to fill in the token fields. Counted with o200k_base (exact for GPT-5 and GPT-4o, a guide for other models); the vocabulary, about 1 MB, downloads when you click.

Currency

Example rate; enter today's rate. Nothing is fetched.

Models to compare

OpenAI
Anthropic
Google
DeepSeek
Alibaba Qwen (Model Studio)
ByteDance Doubao (Volcano Ark)
Moonshot Kimi

Monthly cost

30,000 requests a month: from $10.50 with Claude Haiku 5.5 to $210 with GPT-6.2 Sol.

Monthly API cost by model, cheapest first
ModelPer monthPer 1,000 requests
Claude Haiku 5.5Anthropic · in $0.10 · cached $0.01 · out $0.50 per 1M $10.50 $0.35
DeepSeek FlashDeepSeek · in ¥2.00 · cached ¥0.04 · out ¥8.00 per 1Mpeak hours · ¥186 in its own currency $26.20 $0.8732
GPT-5 miniOpenAI · in $0.25 · cached $0.025 · out $2.00 per 1M $35.25 $1.18
Gemini 3.5 Flash-LiteGoogle · in $0.30 · cached $0.03 · out $2.50 per 1M $43.50 $1.45
Claude Sonnet 5.5Anthropic · in $2.00 · cached $0.10 · out $10.00 per 1M $210 $7.00
GPT-6.2 SolOpenAI · in $2.00 · cached $0.10 · out $10.00 per 1M $210 $7.00

List prices checked 2026-11-18 on each provider's pricing page (click a price to open it). Not included: cache write fees, batch and off-peak discounts unless listed, regional surcharges, images, audio and tool fees, taxes.

Share these inputs: the values go after the # in the link, so they never reach a server.

Everything is calculated in your browser. The share link keeps only the numbers, currency and model names after the # sign, which browsers do not send to servers; sample text is never put in the link.

Version v1.0Updated Update history

Read the result

What the table shows

One row per ticked model, cheapest first, all converted to the currency you picked.

Per month

Requests per day × days, times the cost of one request: uncached input, cached input and output, each at its own price.

Per 1,000 requests

Handy for comparing with a per-conversation budget or with what a support ticket costs today.

Price line

The provider's list price per million tokens in its own currency. Click it to open the pricing page the number came from.

Notes

Long-prompt tiers, peak and off-peak prices, and models without a separate cache price are flagged on the row.

Output usually costs 4 to 8 times as much as input, so the length of answers often matters more than the length of prompts.

How to use it

Estimate the cost in four steps

Start rough; the ranking rarely changes when the numbers move a little.

  1. 1

    Count your requests

    From logs, or from expected conversations × messages per conversation. Each answer is one request.

  2. 2

    Measure the tokens

    Paste a real prompt and answer into the sample box, or use the token counter. Include the system prompt and history.

  3. 3

    Set caching and currency

    If the same long system prompt or documents are sent every time, a cache hit rate of 50% or more is common.

  4. 4

    Tick models and compare

    Pick the models you would consider; share the link with your team.

Formula

How the monthly cost is calculated

The same arithmetic the providers use, with list prices per million tokens.

  • Requests per month = requests per day × days per month
  • Cached input = input tokens × cache hit rate; the rest is uncached input
  • Cost per request = (uncached input × input price + cached input × cache price + output × output price) ÷ 1,000,000
  • Tiered prices: when one request's input passes the threshold (for example 272K tokens for GPT-6 models), the whole request uses the higher tier
  • USD prices × exchange rate = CNY; CNY prices ÷ exchange rate = USD

Cost per request

(in × (1 − hit) × Pin + in × hit × Pcache + out × Pout) ÷ 106

P = list price per million tokens

Monthly cost

cost per request × requests per day × days

Converted with the exchange rate you enter

Worked example

A support bot answering 1,000 questions a day

Each call sends 1,500 input tokens (system prompt, help article, question) and gets a 400-token answer, 30 days a month.

Monthly cost of the same workload on two models
ModelCost per requestPer month
GPT-6.2 Sol ($2 in, $10 out per 1M)1,500 × 2 + 400 × 10 = 7,000 ÷ 10⁶ = $0.007× 30,000 = $210
DeepSeek Flash, peak (¥2 in, ¥8 out per 1M)1,500 × 2 + 400 × 8 = 6,200 ÷ 10⁶ = ¥0.0062× 30,000 = ¥186 (about $26.2 at 7.1)
DeepSeek Flash with 50% cache hits (¥0.04 cached)750 × 2 + 750 × 0.04 + 400 × 8 = 4,730 ÷ 10⁶ = ¥0.0047× 30,000 = ¥141.9

Enter 1000, 1500, 400 and 30 in the calculator and tick these models to get the same numbers.

Common mistakes

Why real bills come out higher

The per-token price is rarely the problem; the token count is.

Forgetting the history

In a chat, every new message resends the earlier turns. Input per request grows with conversation length.

Ignoring reasoning tokens

Reasoning models bill their hidden thinking as output tokens, often several times the visible answer.

Comparing ChatGPT plans with API prices

ChatGPT Plus and similar subscriptions are per person; the API is billed per token. They are different products.

Using stale prices

Providers change prices and add tiers often. Check the date under the table and the pricing page before committing a budget.

FAQ

LLM cost calculator questions

Still have a question?

Chat with the team behind TapCub — we usually reply within a few hours on business days.

How much does the OpenAI API cost per request?

It depends on the model and the tokens. With GPT-6.2 Sol at $2 per million input and $10 per million output tokens, a request with 1,500 input and 400 output tokens costs about $0.007. Enter your own numbers above to compare models.

Is ChatGPT Plus the same as API pricing?

No. ChatGPT plans are monthly subscriptions for people using the ChatGPT app. The API, which apps and chatbots call, is billed per token at the prices in this calculator.

How do I estimate tokens per request?

Paste a real prompt and answer into the sample box and click count, or use the token counter. Include the system prompt, chat history and any documents you add. As a guide, 1,000 English words are roughly 1,200 to 1,400 tokens with OpenAI tokenizers.

What is the cache hit rate?

Most providers charge much less for input they have seen recently, such as a long system prompt that starts every request. If that fixed part is 60% of your input and calls are frequent, use about 60%. Some providers also charge to write the cache, which is not included here.

Why are Chinese models priced in yuan?

DeepSeek, Alibaba Qwen, Doubao and Kimi publish yuan prices for their China platforms. The calculator keeps the original price and converts with the exchange rate you enter, so you can match it to your invoice currency.

How current are the prices?

Every price was checked on 2026-11-18 on the provider's own pricing page, linked from each row. Models whose price could not be confirmed, or that have an announced price change, are left out.

Are batch or off-peak discounts included?

Batch APIs (usually 50% off) are not included. DeepSeek peak and off-peak prices are listed as separate rows because they apply at different times of day.

Changelog

Update history

New features, fixes and data refreshes for this tool, newest first.

  1. · v1.0

    LaunchedFirst release.

Data this tool relies on

  • LLM API list prices, tokenizer ratios and context windows — checked 2026-11-18, reviewed every 1 months (next review by 2026-12-18)

Missing something or found an error? Tell us — requests that come up often go into the next update. All tool updates

Next step

Answer chat questions with AI on a fixed plan

TapCub live chat answers from your knowledge base with the source named, hands off when unsure, and includes AI replies in every plan.

See your data clearly. Find your growth.

Every click, backed by data. Install one line of code and see your first numbers in a minute.

No credit card · Free plans for analytics and chat · Cookieless analytics