LLM Cost Estimator
Estimate multi-model API cost from tokens or pasted prompts.
OpenGet a quick token count and rough embedding cost for any block of text.
Use a passage the size of a real chunk.
The tool estimates tokens at about four characters each.
A reference-rate cost appears with the rate shown.
Multiply by your passage count for a corpus estimate.
You are about to embed a knowledge base for search and want a rough sense of scale before kicking off a long batch job. Exact numbers can wait; the order of magnitude cannot. One representative passage pasted here is enough to get it.
Paste a representative passage and the output shows three lines: approximate tokens at four characters each, a typical vector dimension of 1536 for reference, and an estimated cost at a fixed reference rate per million tokens printed in the result. Multiply the figure by the number of similar passages in your corpus to project the total.
The rate is a single built-in value, not your provider's current price, so check your vendor's pricing page and scale accordingly. A more useful habit is to paste a realistic chunk rather than a short sentence, because chunk size drives the total far more than per-token price. For a fuller picture that includes overlap, chunk count and a price you type yourself, use the RAG Chunk Simulator. The token cost comparison guide explains why embedding and chat token counts can differ, and a word-based token estimate gives you a second opinion.
The arithmetic runs in this tab. Nothing is embedded or sent anywhere, so it is fine to paste text from private documents.
Keyboard shortcut
Answers for this tool. For site-wide help, open the FAQ hub.
Not here. The rate is fixed and shown in the output so you can rescale it. If you want to type your own price, the RAG Chunk Simulator has a price field and a document count.
It is a common vector size for text embedding models, shown for orientation only. Your model may produce more or fewer dimensions, which affects storage in the vector database but not the token count.
Yes, in a real pipeline. Overlap repeats text at chunk boundaries, so you embed more tokens than the source contains. This tool only measures the text you paste, so add your overlap percentage on top.