Count tokens in your text and estimate input costs across the models in the table.
Use a representative prompt or document. Include the instructions and context you expect to send with it.
The tokenizer runs in your browser. Words, characters, and tokens are different measures.
Other models may use different tokenizers. The table also excludes reply tokens and any additional request content.
No. The counter uses one tokenizer; other providers can count the same text differently. Use the relevant provider’s tokenizer for an exact request budget.
No. Leave room for instructions, tools, and the reply, and test whether the model uses the relevant parts of the text correctly.
The browser counts text with the installed gpt-tokenizer encoding, then estimates input cost using the reviewed price table.
Illustrative example: If the counter estimates 1,000 tokens and the selected input price is $2 per million, the estimated input cost is $0.002.
Tokenisation differs by model and language. Treat the count as an estimate for this comparison, and use the provider’s tokenizer or returned usage for billing.
Bring the result and one workflow to a free 30-minute call. We will review the assumptions and discuss a practical next step.