Token counter for GPT, Claude and Gemini

How much will this request cost?

Input cost is your input token count multiplied by the model’s published input price. Output cost uses the expected output length you enter; it is an assumption, not a prediction. Cached input is priced separately where the provider publishes a cached-input rate, and the request total is multiplied by requests per month when you turn that option on. Together, this is an LLM cost calculator for a transparent text-only scenario.

The default example is 1,000 input tokens and 500 output tokens. On GPT-6.1 Sol at the checked rate, that is $0.0070 per request. Real bills can also include system messages, tools, images, files, long-context or other provider-specific charges that this small calculator does not model.

Prices are API list prices in USD per 1 million tokens, before tax. Last checked: November 11, 2026. Claude API pricing and plan limits. OpenAI tokenizer, OpenAI API pricing, Anthropic token counting, Anthropic API pricing, Gemini token counting, Gemini API pricing.

What is a token?

A token is a piece of text a model reads and bills for. It may be a whole word, part of a word, punctuation, a code fragment or an emoji. There is no universal words-to-tokens conversion, so paste representative text instead of relying on a fixed ratio.

Why do providers show different counts?

Each provider uses its own tokenizer. A real request can also add system instructions, tools, images, files and generated output. This page keeps those pieces visible: the input count is separate from your expected output and cached-input assumptions.

What is exact here?

OpenAI uses an o200k-family tokenizer in your browser and is labelled OpenAI-compatible because model-specific request formatting can still add tokens. Claude and Gemini use their official count service when Bento has the provider key; if that service is not configured, fails or is rate-limited, the card shows a labelled estimate instead of a blank result.

How should you use the result?

Budget with a representative prompt, a realistic expected reply length and the number of requests you expect. Then compare the estimate with the provider’s usage report after shipping; the provider’s report is the billable source of truth.

Does the token counter send your text anywhere?

OpenAI-compatible counts run in your browser. When you select Claude or Gemini and an official count is configured, this page sends your text through Bento's server to that provider only to count tokens; otherwise it stays in the browser as a character-based estimate. We do not store it, put it in a URL, send it to analytics, or show provider keys in the browser. Read more about what providers do with API prompts.

Make the finished thing

Bento helps you make finished writing, images, video, voice, apps, 3D and games in one subscription from $20/month. It picks the best model for each task and switches as better models arrive, so you do not have to juggle separate provider accounts. This calculator estimates provider API bills; it is not a Bento-credit meter. See Bento's plans. 7-day free trial.

FAQ

How many tokens are in my prompt?

Paste it into the counter. OpenAI uses a browser tokenizer; Claude and Gemini request their official count when the server-side provider key is available, and otherwise show a labelled estimate.

Are these token prices for API calls or subscriptions?

API calls. The prices are per 1 million input, output and cached-input tokens. They are not ChatGPT Plus, Claude Pro or other subscription prices.

Why is output cost an estimate?

The prompt tells us the input length, not how long the model will answer. Enter an expected output length so the calculator can show a transparent scenario rather than pretend it knows the future response.

Does this calculate the complete provider bill?

No. It covers text input, expected text output and cached input. Tools, images, files, system messages, long-context pricing, batch mode and other provider-specific charges can change the bill.

Are Claude and Gemini counts private?

The official count request sends the text through Bento's server to the selected provider. Bento does not store the text or send it to analytics; the provider receives it to count tokens. When the official request is unavailable, the page uses a local character-based estimate instead.