Exact counts, honest labels
Counts use the o200k tokenizer family — exact for OpenAI models and a close, clearly-labeled approximation for Claude, Gemini, Grok, Kimi, and DeepSeek, whose tokenizers are not public.
Prompt calculator
Paste a prompt. Get its real token count and price.
The prompt calculator counts tokens with a real BPE tokenizer that runs in your browser, then prices the same prompt across every audited model — per request and per month.
Free · no sign-upCalculations run in your browser.Pricing registry reviewed Aug 16, 2026
Same prompt, every model
Cost per request and per month
| Model | Per request | Per month | Context used |
|---|---|---|---|
| DeepSeek V4 FlashDeepSeek | $0.0002 | $1.07 | <0.1% of 1.05M |
| Llama 4 MaverickMeta | $0.0005 | $2.96 | <0.1% of 1M |
| DeepSeek V4 ProDeepSeek | $0.0006 | $3.31 | <0.1% of 1.05M |
| GPT-5.6 LunaOpenAI | $0.0007 | $4.40 | <0.1% of 1.05M |
| Mistral Large 3Mistral | $0.0009 | $5.60 | <0.1% of 262K |
| Gemini 3.5 Flash-LiteGoogle | $0.0015 | $9.12 | <0.1% of 1.05M |
| Gemini 3.6 FlashGoogle | $0.0023 | $13.81 | <0.1% of 1.05M |
| Gemini 3.7 FlashGoogle | $0.0023 | $13.81 | <0.1% of 1.05M |
Planning estimates from the PromptCostLab registry (re-verified Aug 16, 2026). Request-level fees, batch discounts, and regional surcharges are excluded; long-context tiers are applied automatically.
One prompt, every price
The comparison table applies your prompt, output allowance, and daily volume to every model in the registry, including long-context tiers where they apply.
Private by default
Your prompt never leaves the browser. The tokenizer loads as a static script; no text is sent to any server.
Plain-English answers
Questions people ask before using the result
How do I calculate the cost of a prompt?
Count the prompt's tokens, add your expected output tokens, multiply each by the model's per-million prices, and multiply by your request volume. This calculator does all of it — across every major model at once.
Is the token count exact?
For OpenAI models, yes — the browser runs the same o200k tokenizer family the models use. For Claude, Gemini, Grok, Kimi, and DeepSeek it is a close approximation, since those providers do not publish their tokenizers.
How many tokens is my prompt?
It depends on length and style: ordinary English runs about 4 characters or 0.75 words per token. Paste your prompt above for a real count instead of a rule of thumb.