Count tokens, then price them across every model.
Paste a prompt to estimate its token count, set your expected response length, and see exactly what one call — or a million — would cost on each model.
Token counts are an estimate (≈ 4 characters per token); actual tokenization varies by model and language. Costs use each model’s standard input and output rates per 1M tokens. Prompt caching would lower the input portion further.
How token counting works
Language models don’t read words — they read tokens, sub-word chunks produced by the model’s tokenizer. In English, one token is about 4 characters or roughly ¾ of a word, so a 1,000-character prompt is on the order of 250 tokens. This tool uses that ratio to give a fast, model-agnostic estimate as you type, then prices the result against every model’s live rates.
Exact counts differ by tokenizer: code, punctuation, non-English text, and rare words all tokenize differently, and each provider ships its own tokenizer. For budgeting, the estimate here is close enough to rank models and size a bill; for exact accounting, use the provider’s own tokenizer or API usage response.
Why the cheapest model depends on your prompt
Input and output are priced separately, and models have very different input-to-output ratios. A long prompt with a short answer is dominated by the input rate; a short prompt with a long answer is dominated by the output rate. That’s why the ranking above re-sorts when you change the expected output length — the “cheapest” model for your workload isn’t always the one with the lowest headline price — see Input vs Output Tokens for why. To model steady monthly traffic instead of a single call, use the cost calculator; to find the lowest-cost models overall, see cheapest models.