Hong Life CalculatorsHow it works
Chunks per document = 1 if the document fits in one chunk, otherwise ⌈(document tokens − overlap) ÷ (chunk size − overlap)⌉. Tokens embedded per document = document tokens + (chunks − 1) × overlap, because overlapping text is embedded twice. Cost = tokens × embedding price ÷ 1,000,000. Raw storage = vectors × dimensions × bytes per dimension (1 GB = 10⁹ bytes); indexes, metadata and replicas add to this, so check your vector database’s sizing guide. Queries are embedded too: queries per day × 30 × tokens per query.
Embedding prices in our price file (US$ per 1M tokens, checked 10 Nov 2026)
| Provider | AI model | Price |
|---|---|---|
| gemini-embedding-001 | 0.15 | |
| Mistral AI | Mistral Embed | 0.1 |
⚠ Estimates for information only — not financial advice.
FAQ
Which chunk size should I use?
There is no single answer. 300–1,000 tokens with a small overlap is a common starting point; test retrieval quality on your own questions and adjust.
Does a smaller number format lose accuracy?
float16 and int8 halve or quarter the storage and usually lose a little retrieval accuracy. Many vector databases offer quantisation; compare results on a sample before switching.
Checked 10 Nov 2026. Embedding prices from our AI price file (Google gemini-embedding-001 $0.15 and Mistral Embed $0.10 per 1M tokens on the official pricing pages, checked 10 Nov 2026); IEEE 754 sizes for float32/float16.