Prompt Token Saver

Paste your prompt and get an item-by-item bloat report — wordy phrases, repeated instructions, filler and politeness overhead — each with estimated tokens wasted. Tick what to cut, generate the leaner version, and compare per-call cost. Estimates, not exact counts.

Advertisement

An honest note about rule-based analysis

This tool matches fixed patterns — it does not understand your intent. A "please" it flags might be required by your style guide; a repeated sentence might be deliberate emphasis; removing "very" can flatten tone. That is why every finding has its own checkbox: you decide what is actually bloat. Always re-read the leaner version before sending it to a paid API, especially for system prompts that run thousands of times.

How the estimates work

Token counts use the same heuristic as our Token Counter: whitespace-separated words times a per-family factor (GPT 1.33, Claude 1.35, Grok 1.33) plus CJK characters times a CJK factor (GPT 1.5, Claude 1.6, Grok 1.5). Real tokenizers split text into subword pieces differently, so treat the numbers as estimates, usually within ~20% for plain English text.

Price sources (verified 2026-10-01)

Claude Sonnet 5 $3/$15 and Opus 4.8 $5/$25 per 1M input/output tokens (Anthropic list prices, widely mirrored); Claude Haiku 4.5 $1/$5; Grok 4.7 $2/$6 (xAI pricing page); GPT-5.5 $5/$30 (OpenAI list price). Costs use input prices only, since this tool trims your prompt. Cached-input, long-context and batch tiers are not included.

FAQ

Are these token counts exact?
No. Each provider uses its own tokenizer and none of them run in a browser page like this. The counts here are heuristic estimates — words times a per-family factor plus CJK characters times a CJK factor — usually within ~20% for plain English text. For billing, always trust your provider's dashboard.
How is this different from Prompt Compressor?
Prompt Compressor applies its rule set in one click and shows the result. Token Saver lists every finding one by one with a checkbox, so you decide what to cut — wordy phrases, filler, politeness overhead, repeated instructions — before generating the leaner version. Use Saver when you want control over each edit, Compressor when you want one click.
Are the cost savings exact?
No. All cost numbers are estimates for reference only: token counts are heuristic and public prices change. Prices on this page were verified on 2026-10-01 — confirm on the provider's official pricing page before budgeting real spend. Nothing here guarantees actual savings.
Does my text leave my browser?
No. Analysis, token math and cost estimates happen entirely in JavaScript on your device. Nothing is uploaded anywhere.
Advertisement