How are token counts estimated?

The tool uses a character-based heuristic of roughly four characters per token, which is close for English prose. It is an estimate, not an exact tokenizer count, but it is consistent across variants so comparisons are fair.

Why does completion token count matter?

Output tokens usually cost several times more than input tokens, so two prompts with identical input cost can differ wildly in total cost if one elicits longer answers. Setting an expected completion length makes the cost-per-call comparison realistic.

Is my prompt text uploaded?

No. All counting and scoring happen locally in your browser. Nothing you paste is sent to a server, stored, or logged.

Prompt Template Cost Optimizer

Q: What is the completeness score?

It is a lightweight heuristic that rewards the presence of common instruction cues such as a role, explicit constraints, output-format directions, and examples. A higher score means the prompt is more likely to be unambiguous, not that it is objectively better.

Prompt template cost optimizer

A prompt that is twice as long costs twice as much on input — and at scale that difference compounds into real money. This tool lets you paste several versions of the same template and compare them head to head on token count, cost per call, projected daily spend, and a rough completeness score, so you can trim verbosity without dropping the instructions that actually matter.

How it works

Each variant’s character length is converted to an estimated token count (about four characters per token for English) and priced at the selected model’s input rate. Your expected completion tokens are priced at the output rate and added, giving a realistic cost per call. Multiplying by your daily call volume projects daily spend. Separately, a completeness heuristic scans for instruction cues — a defined role, explicit constraints, output-format directions, and examples — and scores each variant so you can spot a cheaper prompt that is also under-specified.

Tips and notes

Cheapest is not always best. Use the completeness score as a guardrail: a variant that saves tokens but loses a format instruction may cost you in retries and bad outputs.
Trim the static boilerplate. Long fixed preambles repeat on every call — the biggest savings usually come from compressing them, not the variables.
Measure output length too. If completions are long, switching to a cheaper generation model often beats shaving the prompt.
Estimates, not exact counts. For precise billing, confirm with your provider’s tokenizer; the heuristic here is for fast comparison.