Prompt Token Optimizer

Trim verbose phrases, politeness fillers, and imperative softeners. See token + dollar savings projected across models.

Original prompt…

Optimized prompt

…

to help me with my project, I would really appreciate it if you could take a look at the following code and let me know if you think there might be any issues with it. basically just go through it line by line and now, point out anything that seems like it could potentially be a problem. also explain why each issue is actually an issue? !

Optimization categories

Verbose phrase tightening“in order to” → “to”, “due to the fact that” → “because”, etc. Meaning-preserving.Politeness fillersStrips “please”, “could you”, “thanks in advance”. The model doesn't need pleasantries.Imperative softenersRemoves “I want you to”, “can you”, “I'd like you to”. The model already knows.Hedging / fillerMay shift toneRemoves “perhaps”, “I think”, “kind of”, “actually”. Can shift tone — opt-in.IntensifiersMay shift toneRemoves “very”, “really”, “quite”, “extremely”. Can change emphasis — opt-in.Whitespace + quotesCollapses runs of spaces/blank lines, normalizes smart quotes and dashes to ASCII.

Tokens saved

−0

% reduction

0.0%

Chars saved

112

Rules applied

Cost savings projection

Calls / month

Model	$ / M input tok	Monthly cost (before)	Monthly cost (after)	Monthly savings
Claude Opus 4.6	$15.00	$0	$0	$0
GPT-5	$5.00	$0	$0	$0
Claude Sonnet 4.6	$3.00	$0	$0	$0
Gemini 3 Pro	$2.50	$0	$0	$0
Claude Haiku 4.5	$0.80	$0	$0	$0
GPT-5 mini	$0.30	$0	$0	$0

Token counts use the OpenAI o200k tokenizer; actual counts on other models will differ ±5–10%. Output costs are not included — focus is on input-prompt savings.

What changed (8 edits)

Category	Rule	Times
verbose	"in order to" → "to"	1
verbose	"at this point in time" → "now"	1
polite	Drop "please\|kindly\|if you would\|if you wouldn'?t mind"	2
polite	Drop "would you please \|kindly ?"	1
polite	Drop "if it'?s not too much trouble,?"	1
polite	Drop "thanks in advance\|so much.?"	1
imperative	Drop "I want\|need you to"	1

About this tool

All transformations are local rule-based rewrites — your prompt never leaves the browser. The default-enabled categories (verbose phrases, politeness fillers, imperative softeners, whitespace) are designed to be meaning-preserving. Hedging and intensifier removal can subtly shift the tone of a prompt and are off by default.

This is not paraphrasing. A rule-based optimizer can't restructure long sentences or rewrite paragraphs. For higher reductions you'd need an LLM rewrite step, which would cost as much as you'd save. The biggest wins from this tool come from prompts written conversationally that include a lot of politeness and softening — typical first-draft prompts give up 15–30% of their tokens to filler.

Cost projections assume the prompt is reused across many calls (e.g. a production system prompt or RAG template). Optimizing a one-shot prompt rarely pays back the effort.