Sharma (home)

Free tool

AI Token Calculator

How much does an AI API call cost? Usually a fraction of a cent. For example, a 1,000-token prompt with a 500-token reply on GPT-6.1 Sol ($2 input / $10 output per 1M tokens) costs $0.002 + $0.005 = $0.007, so 10,000 calls a month is about $70. Paste your own prompt below to compare OpenAI, Claude, Gemini, Kimi and DeepSeek.

Estimated input tokens

0

0 characters

Requests

Monthly = per request × requests per day × 30.4

ModelInput / Output per 1MPer requestPer month12 months
GPT-6 AstraOpenAI · flagship · checked Nov 15, 2026$10 / $50$0.0250$0.7600$9.12
GPT-6.1 SolOpenAI · mid · checked Nov 15, 2026$2 / $10$0.00500$0.1520$1.82
GPT-6.1 TerraOpenAI · mid · checked Nov 15, 2026$1 / $5$0.00250$0.0760$0.9120
GPT-6 LunaCheapestBest value: BudgetOpenAI · budget · checked Nov 15, 2026$0.1 / $0.5$0.00025$0.00760$0.0912
GPT-5.4 miniOpenAI · budget · checked Nov 15, 2026$0.75 / $4.5$0.00225$0.0684$0.8208
Claude Opus 5.5Anthropic · flagship · checked Nov 15, 2026 · +30% tokenizer$4 / $20$0.0130$0.3952$4.74
Claude Sonnet 5.5Anthropic · mid · checked Nov 15, 2026 · +30% tokenizer$2 / $10$0.00650$0.1976$2.37
Claude Haiku 5.5Anthropic · budget · checked Nov 15, 2026 · +30% tokenizer$0.1 / $0.5$0.00032$0.00988$0.1186
Gemini 4 ArgonGoogle · flagship · checked Nov 15, 2026 · promo price through Nov 30, 2026$2 / $10$0.00500$0.1520$1.82
Gemini 3.8 FlashBest value: MidGoogle · mid · checked Nov 15, 2026 · promo price through Dec 31, 2026$0.75 / $3.75$0.00187$0.0570$0.6840
Gemini 3.5 Flash-LiteGoogle · budget · checked Nov 15, 2026$0.3 / $2.5$0.00125$0.0380$0.4560
Kimi K3.1 ProKimi · flagship · checked Nov 15, 2026 · batch billed at 60% of standard$2 / $10$0.00500$0.1520$1.82
DeepSeek V4 ProBest value: FlagshipDeepSeek · flagship · checked Nov 15, 2026 · peak rate; off-peak is half$1.32 / $3.96$0.00198$0.0602$0.7223
DeepSeek FlashDeepSeek · budget · checked Nov 15, 2026 · peak rate; off-peak is half$0.3 / $1.2$0.00060$0.0182$0.2189

Want this running in your product?

With GPT-6 Luna: $0.00760/month, $0.0912 over 12 months.

Talk to me about building it →

OpenAI: checked Nov 15, 2026 (source)

Anthropic: checked Nov 15, 2026 (source)

Google: checked Nov 15, 2026 (source)

Kimi: checked Nov 15, 2026 (source)

DeepSeek: checked Nov 15, 2026 (source)

Standard tier, short-context prices. Cache-write fees and long-context surcharges are not included. Token counts are estimates.

How this AI token calculator works

AI models don't read words; they read tokens, small chunks of text produced by a tokenizer. In English one token averages about four characters, or roughly three quarters of a word. This tool blends a character-based estimate with a word-based one, which tracks modern tokenizers closely for normal prose without loading a heavy tokenizer into your browser.

Cost is then simple arithmetic: input tokens × the model's input price, plus output tokens × its output price, divided by one million. Output is usually priced four to eight times higher than input, so a long answer often costs more than a long prompt.

Using it as an OpenAI token calculator

For OpenAI models, the estimate is a good stand-in for the official tokenizer on everyday English. If you're budgeting an app, enter a realistic reply length and the number of requests per day or month to get an OpenAI API pricing calculator view of your bill. Remember that system prompts, tool definitions and chat history are also input tokens on every call.

What the estimate leaves out

  • Cache-write fees. The calculator applies the lower cached-input rate but does not include the one-time write fee some providers charge when the prompt is first cached.
  • Long-context tiers. Some models charge more above a prompt-size threshold (for example 272K tokens on several OpenAI models, 100K on Claude Haiku 5.5).
  • Reasoning tokens. Thinking models bill hidden reasoning as output, so real output can exceed the visible reply.
  • Images, audio and tools. These are priced separately.

For official per-token rates, see Claude API pricing, OpenAI API pricing, Gemini API pricing, and Kimi and DeepSeek pricing.

If you're comparing tools rather than raw APIs, see my breakdowns of Claude Code pricing, Cursor pricing and ChatGPT vs Claude, plus Claude vs Gemini and ChatGPT vs Copilot. Planning a product that uses AI? The website cost calculator helps you estimate the build.

FAQ

How accurate is this AI token calculator?
It is an estimate. For typical English text it usually lands within about 10–15% of the real count. Code, non-English text and unusual formatting can differ more. For an exact number, use each provider's token counting endpoint.
How many tokens is 1,000 words?
Roughly 1,300–1,400 tokens in English, since one token averages about four characters or three quarters of a word.
Does the OpenAI token calculator include output tokens?
Yes. Enter the expected response length in tokens and the tool adds output cost, which is usually several times more expensive per token than input.
Why are Claude counts higher?
Anthropic says Claude 4.7 and later models use a newer tokenizer that produces roughly 30% more tokens for the same text, so the tool applies that adjustment to Claude models.
Is my text sent anywhere?
No. Everything runs in your browser. Nothing you paste is uploaded or stored.
How does prompt caching change the cost?
When part of your prompt repeats across calls, such as a long system prompt, providers bill the cached part at a lower rate. On Claude Opus 5.5, for example, cached input is $0.20 per 1M tokens against $4 for normal input. Turn on caching in the calculator and set the share of input you expect to be cached. Cache-write fees are not included.
How much does the Batch API save?
OpenAI, Anthropic and Google take 50% off input and output for batch jobs that can wait for results. Kimi bills batch jobs on Kimi K3.1 Pro at 60% of the standard price. DeepSeek does not list a batch API, so the batch toggle leaves its rows unchanged.
How do I budget monthly AI API costs?
Multiply the cost per request by your daily requests and by 30.4, the average days in a month. Count the system prompt and chat history on every call, since they are input tokens too, and add a margin for retries and longer replies.