01
Unix Timestamp Converter
Converts Unix epoch timestamps to human-readable dates in any time zone, and dates back to…
Embedding Cost Estimator
Estimates what embedding a corpus costs, how many chunks it produces, and how much vector storage it needs.
Developer
Nothing uploaded
No sign-up
Free, unlimited
Blank uses the preset — override it, presets go stale.
Leave at 0 for the preset.
Enter your figures and press Calculate.
In short
Embedding is almost never the expensive part of a retrieval system, and it is the part people budget for. At current prices a corpus of a hundred thousand documents costs a few dollars to embed once. The real costs come afterwards: the vector database, and above all the generation calls that use the retrieved text, which are typically orders of magnitude larger. Two things do catch people out — chunk overlap means embedding more text than you actually have, so a 20% overlap costs 25% more than the raw corpus; and changing embedding model means re-embedding everything, because vectors from two different models are not comparable and there is no incremental migration.
How to use the Embedding Cost Estimator
- Enter how many documents you have and their average length.
- Set the chunk size and overlap you plan to use.
- Correct the price from the provider's page.
- Compare the one-off cost against the generation costs it enables.
About the Embedding Cost Estimator
The Embedding Cost Estimator estimates what embedding a corpus costs, how many chunks it produces, and how much vector storage it needs. It runs as plain JavaScript inside the page, which means there is no upload step, no queue and no size limit imposed by a server — the only limit is your own device.
What people use it for
- Costing a retrieval system before building it.
- Sizing a vector database.
- Working out what re-chunking would cost to redo.
Details
| Embedding | Usually the cheapest part of a retrieval system |
|---|---|
| Overlap | A 20% overlap embeds 25% more text than you have |
| Changing model | Requires re-embedding everything — vectors are not comparable |
| Storage | Dimensions × 4 bytes per vector, plus index overhead |
The maths behind it
This tool uses chunks = total tokens ÷ (chunk size × (1 − overlap)); tokens embedded = chunks × chunk size; cost = tokens ÷ 1,000,000 × price. Every intermediate value is shown on screen, so the result can be checked by hand.
The storage figure assumes float32 vectors plus about 40% index overhead. Quantising to int8 cuts that by roughly three quarters with a small accuracy cost, and is worth considering well before a corpus gets large.
Where this tool stops
Estimates from averages. Real corpora have a long tail of very long and very short documents, and chunking on structure rather than a fixed size changes the count materially — usually upward.
Frequently asked questions
Why does overlap cost more than I expected?
Because it is measured against the chunk, not the corpus. A 500-token chunk with 20% overlap advances only 400 tokens per chunk, so you need 25% more chunks — and each is still 500 tokens to embed. The arithmetic is easy to get backwards and the error is always in the expensive direction.
What happens if I change embedding model?
You re-embed everything. Vectors from two different models occupy different spaces and cannot be compared, so there is no way to migrate part of an index — which makes the choice of model a bigger commitment than its price suggests, and a good reason to test on a sample first.
Is the vector database or the embedding the bigger cost?
Almost always the database, and then the generation calls that follow. Embedding a large corpus is a one-off in single or double-digit dollars; storing and serving the vectors is monthly, and the LLM calls that use what is retrieved usually dwarf both.
Does the Embedding Cost Estimator cost anything?
No charge and no sign-up. There is no free tier to run out of because there is no paid tier — inputtesthub.com is funded by advertising rather than by metering the utility.
Is my data uploaded anywhere?
No. The Embedding Cost Estimator runs entirely in your browser using JavaScript. Nothing is transmitted, logged or stored, which you can verify yourself in the Network tab of your browser's developer tools.
Can I check the maths myself?
It uses chunks = total tokens ÷ (chunk size × (1 − overlap)); tokens embedded = chunks × chunk size; cost = tokens ÷ 1,000,000 × price. The full working is printed under the result, so you can follow every step rather than trusting a black box.
When would I actually need the Embedding Cost Estimator?
In practice it gets used for three things: costing a retrieval system before building it. Sizing a vector database. Working out what re-chunking would cost to redo. It is built for exactly those cases rather than being a general-purpose tool with this bolted on.
What are the limits of the Embedding Cost Estimator?
Estimates from averages. Real corpora have a long tail of very long and very short documents, and chunking on structure rather than a fixed size changes the count materially — usually upward. Knowing where a tool stops being trustworthy matters more than the cases where it works.
What are the reference figures here?
Embedding is Usually the cheapest part of a retrieval system. Overlap is A 20% overlap embeds 25% more text than you have. Changing model is Requires re-embedding everything — vectors are not comparable. These are printed on the page itself, so you do not have to take them on trust.
Is the number rounded?
The calculation runs in double precision and the result is shown to ten significant figures, so converting a value and converting it back returns the original. What is displayed is rounded for readability; what is copied is not.
Related developer tools
All 9202
Subnet Calculator
Calculates network address, broadcast, usable host range, wildcard mask and host count from an…
03
IP Address Converter
Converts IPv4 addresses between dotted decimal, integer, binary and hexadecimal notation.
04
Semantic Version Comparator
Compares two semantic versions and says which is newer, what level of change it is, and…
05
Compression Ratio Calculator
Converts between compression ratio, percentage saved and bits per byte, and works out the…
06
DNS Propagation Time Calculator
Works out how long a DNS change takes to be seen everywhere, from the TTL that was published…
07
LLM API Cost Calculator
Works out what an LLM feature costs per call, month and year, including what prompt caching…
08
Context Window Planner
Works out how many conversation turns fit in a context window once the system prompt, tools…
09
JSON Formatter
Formats, indents and validates JSON with precise error positions, key sorting and a…
10
JSON Validator
Checks whether JSON is valid and reports the exact line, column and reason for the first error…