Embed Cost Estimator

A lightweight Python library and CLI to estimate OpenAI embedding costs.

Installation

Install from PyPI:

pip install embedding-cost-estimator

Basic CLI Usage (Rough Estimate)

Run a quick rough estimate using a simple chars/4 heuristic:

embed-cost --chunks <NUM_CHUNKS> --chars <AVG_CHARS_PER_CHUNK> [--model <MODEL>]

#--chunks, -n  Number of chunks (required)

#--chars, -c  Average characters per chunk (default: 500)

#--model, -m  Embedding model choice (default: text-embedding-ada-002)

CLI Options

Examples:

1. Default model, custom sizes

embed-cost --chunks 1000 --chars 500
#Estimated embedding cost: $0.050000

2. Using a different model

embed-cost --chunks 500 --chars 300 --model text-embedding-3-small
# Estimated embedding cost: $0.003000

Python API

You can call estimate_embedding_cost() in two mutually-exclusive ways:

1. Rough estimate

Rough estimate using a simple chars/4 heuristic

from embed_cost import estimate_embedding_cost

cost = estimate_embedding_cost(
    num_chunks=250,
    chunk_size_chars=400,
    model="text-embedding-3-small",
)

print(f"Rough cost: ${cost:.6f}")

2. Precise mode (exact token counts via tiktoken):

For exact token counts via tiktoken, by passing your list of text chunks

from embed_cost import estimate_embedding_cost

# your pre-chunked list of text segments
chunked_docs = [
    "First chunk of text…",
    "Second chunk of text…",
    # …etc…
]

cost = estimate_embedding_cost(
    chunk_texts=chunked_docs,
    model="text-embedding-ada-002",
)
print(f"Precise cost: ${cost:.6f}")

Example

1. Exact Token Count in Code

from embed_cost import estimate_embedding_cost

# assuming your document is already split:
chunked = ["Lorem ipsum…", "Dolor sit amet…", …]
cost = estimate_embedding_cost(
    chunk_texts=chunked,
)
print(cost)  # e.g. 0.000320

Contributing

We welcome contributions!

1. Fork the repo and create a feature branch.

2. Run tests and lint locally:

poetry install            # or pip install -e .
poetry run pytest -q      # or pytest -q
poetry run flake8 src tests
poetry run black --check .

3. Open a pull request against main.

4. Maintain 100% test coverage for new code and adhere to Black/Flake8 style.

Please see CONTRIBUTING.md for more details.

License

MIT © Pragasen Naicker