Paste your prompt, set an expected reply length, and instantly compare pricing across GPT, Claude, Gemini, Grok, DeepSeek and Llama.
| Model | Input | Output | Total / request | Per 1,000 requests |
|---|
How this is estimated: token counts are approximated from your text (~4 characters or ~0.75 words per token — the same rule of thumb OpenAI's own docs use), then adjusted per model family. This is not an exact tokenizer, so treat numbers as a close estimate for budgeting — always confirm exact usage in your provider's dashboard before relying on it for billing. Pricing is manually verified against each provider's public pricing page and may change — check the source link for the latest rate before making a purchasing decision.
Almost everyone who is surprised by an API bill made the same mistake: they priced one message, when they were being charged for a conversation.
Models do not read words. They read tokens — fragments that average roughly four characters in English. "Understanding" is one word but several tokens; a comma is a token; a line break is a token. Code, unusual names and non-English text all tokenise less efficiently, which is why the same paragraph in Urdu or Arabic can cost noticeably more than in English.
Providers charge separately for what you send and what the model writes back, and output is typically several times more expensive than input. That single fact changes how you should design a request: a long prompt that produces a short answer is usually cheap, while a short prompt that asks for a long document is not.
A model has no memory between calls. To continue a conversation, the entire history has to be sent again with every new message. Turn one sends your question. Turn ten sends your question, the model's answer, and every exchange since — then charges you for all of it as input.
A chat that feels like ten short messages can bill like a document, because the input side grows with roughly the square of the conversation length. This is the number this calculator exists to make visible before you build something around it.
Nothing you type into this calculator leaves your browser. The token counting and the arithmetic both run on your own device.
Quick answers about how this tool works.
It is a close estimate, not an exact provider count. The calculator uses a blended characters-and-words approximation with a per-family adjustment, which lands near the real tokenizer for ordinary prose. Code, unusual symbols and non-English text can drift further, so treat the number as a good planning figure rather than a billing figure.
Each model's input and output price is taken from that provider's own public pricing page and checked by hand. Prices do change, so always confirm against the provider before committing to a large budget.
No. The whole calculator runs in your browser. Your prompt text never leaves your device and is not sent to any server.
Most providers charge more per output token because generating text is more computationally expensive than reading it. That is why the expected reply length setting matters so much for the final cost estimate.
Background reading on how this actually works.
GPT vs Claude pricing, side by side LLM API pricing explained: what you're actually paying for How to write shorter AI prompts that get the same answer