🧬 Embedding Cost Calculator

Estimate what it costs to embed your document or RAG corpus across OpenAI, Cohere, Google, and open-source models — plus the resulting vector storage footprint.

🧬 Corpus Details
*Open-source models have no per-token API charge, but you still pay for GPU/CPU infrastructure to run them.
Used only for the vector storage size estimate below.
📈 Embedding Cost & Storage
Total Embedding Cost
Estimated Vector Storage
Total Tokens Processed
Cost per 1,000 Chunks
Storage (MB)
Storage (GB)

📊 Cost by Embedding Model (same corpus)

Total Cost by Model
Storage Size by Dimension
⚠️ Prices updated June 2026 — verify current rates on each provider's pricing page. Storage estimate assumes 4 bytes per dimension (float32) and does not include vector database index overhead, metadata, or replication, which typically add 20-50% more on top of raw vector size.
📊

Enter your details and click Calculate to see results

Guide

About the Embedding Cost Calculator

Last updated: August 2026 · Reviewed by the NeftCal editorial team

Before a retrieval-augmented generation (RAG) system, semantic search index, or recommendation engine can retrieve anything, every chunk of your corpus first has to be converted into a numeric embedding vector — and that conversion is billed per token, just like a chat completion. This embedding cost calculator turns your corpus size (chunk count and average tokens per chunk) into a total embedding cost across five common providers, plus the vector storage footprint those embeddings will occupy once they land in a vector database, so you can budget a RAG pipeline's data-ingestion side before you commit to a model.

What This Calculator Estimates

Two numbers, from one set of inputs: the dollar cost to run your entire corpus through an embedding API, and the raw disk space the resulting vectors will need once you pick a dimension. Enter your chunk count, average tokens per chunk, chosen embedding model, and target dimension, and the calculator returns total embedding cost, a normalized cost per 1,000 chunks, total tokens processed, and storage size in both megabytes and gigabytes — plus a side-by-side comparison across five models and five dimensions so you're not stuck evaluating one option at a time.

Who Should Use This Calculator

It's built for RAG engineers scoping a new knowledge-base or document-search feature, ML engineers deciding whether a model or dimension upgrade is worth the re-embedding bill, data platform teams sizing a vector database before provisioning it, startup founders budgeting an AI search feature's ingestion cost, and students or researchers learning how embedding pricing and vector storage scale with corpus size.

Why Embedding and Vector Storage Cost Matters

Embedding cost is often small per token, but multiplying by a corpus of tens or hundreds of millions of tokens turns fractions of a cent into a meaningful line item, especially when a corpus needs periodic re-embedding as documents change. Just as important, and easy to overlook, is the storage side: every embedding vector you generate lives in a vector database indefinitely, so the dimension you choose sets an ongoing infrastructure cost, not just a one-time API charge. Sizing both sides together, the way you would budget a traditional dataset with a Cloud Storage Cost Calculator, helps you avoid picking a high-dimension model for marginal quality gains you didn't actually need.

Real-World Applications

  • Sizing a knowledge-base RAG system's one-time embedding bill before building the retrieval pipeline
  • Budgeting the recurring cost of re-embedding a corpus after switching to a newer or larger embedding model
  • Choosing a chunk size and overlap strategy that balances retrieval quality against total token volume
  • Comparing embedding providers side by side, then pairing with the AI Token Calculator to budget the generation side of a full RAG stack
  • Projecting vector database storage growth alongside a Cloud Storage Cost Calculator estimate for the underlying infrastructure

Tips for Accurate Results

  • Measure your actual average tokens per chunk from a representative sample instead of guessing — chunking strategy changes this number significantly
  • Include chunk overlap in your token estimate; overlapping chunks embed some text twice, which increases both cost and chunk count
  • Remember that switching embedding models or dimensions later generally means re-embedding the entire corpus from scratch
  • Add 20-50% on top of the raw storage estimate here for vector database index structures, metadata, and replication overhead
  • For self-hosted open-source models, budget GPU or CPU compute time separately — the $0 API figure here does not include infrastructure
Formula

The Embedding Cost & Vector Storage Formula, Explained

How this embedding cost calculator turns corpus size into a dollar figure and a storage footprint

Token & Cost Formula
Total Tokens = Chunks × Avg Tokens per Chunk
Total Embedding Cost = (Total Tokens ÷ 1,000,000) × Price per 1M Tokens
Cost per 1,000 Chunks = (1,000 × Avg Tokens per Chunk ÷ 1,000,000) × Price per 1M Tokens

Vector Storage Formula
Storage (bytes) = Chunks × Dimension × 4 bytes
Storage (MB) = Bytes ÷ 1,048,576  |  Storage (GB) = Bytes ÷ 1,073,741,824

Price per 1M Tokens is each embedding model's published rate, shown directly in the model dropdown — from $0.02 for OpenAI text-embedding-3-small up to $0.15 for Google gemini-embedding, or $0 API cost for self-hosted open-source models. The 4-byte figure in the storage formula reflects a standard 32-bit float per dimension, the default precision most vector databases use unless you explicitly quantize to int8 or a lower precision.

📐

Embedding Dimensions

The length of the numeric vector a model outputs — 384 to 3072 in this calculator. Higher dimensions can capture more semantic nuance but cost more to store and, for some providers, more to generate.

✂️

Chunk Size & Overlap

How you split documents into embeddable pieces. Smaller chunks with overlap improve retrieval precision but increase total chunk count — and therefore total tokens billed — for the same source corpus.

💾

Vector Storage Bytes-per-Vector

Each vector needs dimension × 4 bytes of raw storage (float32). A 1,536-dimension vector needs 6,144 bytes; a 3,072-dimension vector needs 12,288 bytes — before any database index overhead.

⚙️ Why This Formula Works

Both halves of the formula are linear multiplications, which mirrors how embedding APIs actually bill (a flat rate per token, the same as chat completions) and how vector storage actually works (a fixed number of bytes per dimension, repeated once per stored vector). That linearity is what makes the cost and storage figures scale predictably as your corpus grows — doubling chunk count roughly doubles both cost and storage.

🎯 When to Use It

  • Before running a first-time embedding job on a new corpus
  • Before deciding whether a dimension or model upgrade is worth the re-embedding cost
  • When sizing a vector database's storage budget ahead of provisioning it

📋 Assumptions

  • Average tokens per chunk is representative of your whole corpus, not just a sample
  • Vectors are stored as standard 32-bit floats, with no quantization applied
  • Pricing reflects each provider's standard published per-token rate

⚠️ Limitations of the Formula

  • Doesn't include vector database index overhead, metadata, or replication (typically +20-50% on raw storage)
  • Doesn't model quantized storage (int8/binary embeddings), which can cut storage 4-32x
  • Doesn't account for provider batch size or rate limits, which affect job duration, not cost
  • Self-hosted models show $0 API cost since actual cost depends on your own GPU/CPU infrastructure
Walkthrough

Step-by-Step: How to Use the Embedding Cost Calculator

From corpus size to a total embedding cost and storage estimate

Enter your number of chunks or documents

Enter the total count of individual text pieces you plan to send to the embedding API — this is your chunk count after splitting documents, not your raw file count.

Set average tokens per chunk

Enter the average token length of one chunk, based on your actual chunking strategy (sentence, paragraph, or fixed token count) rather than a rough guess.

Choose an embedding model

Select from OpenAI text-embedding-3-small, text-embedding-3-large, Cohere embed-v4, Google gemini-embedding, or a self-hosted open-source option — each shows its price per 1M tokens right in the dropdown.

Choose an embedding dimension

Select 384, 768, 1024, 1536, or 3072 dimensions. This only affects the vector storage estimate below, not the embedding cost itself.

Click "Calculate Cost"

See total embedding cost, cost per 1,000 chunks, total tokens processed, and estimated vector storage in MB and GB.

Compare models and dimensions

Review the model comparison list and the two charts to see how switching embedding models or dimensions changes both cost and storage footprint for the same corpus.

Example

Worked Example

Using the calculator's own default scenario — 50,000 chunks, 300 tokens/chunk, OpenAI text-embedding-3-small, 1,536 dimensions

Scenario

Suppose you're embedding a support-docs knowledge base of 50,000 chunks, averaging 300 tokens per chunk, using OpenAI's text-embedding-3-small ($0.02 per 1M tokens) at 1,536 dimensions.

Chunks50,000
Tokens / Chunk300
Modeltext-embedding-3-small
Dimension1,536
Step 1 — Total tokens processed: 50,000 × 300 = 15,000,000 tokens.
Step 2 — Total embedding cost: (15,000,000 ÷ 1,000,000) × $0.02 = 15 × $0.02 = $0.3000.
Step 3 — Cost per 1,000 chunks: (1,000 × 300 ÷ 1,000,000) × $0.02 = 0.3 × $0.02 = $0.006, shown as <$0.01 since it rounds under a cent.
Step 4 — Vector storage in bytes: 50,000 × 1,536 × 4 = 307,200,000 bytes.
Step 5 — Convert to MB: 307,200,000 ÷ 1,048,576 ≈ 293.0 MB.
Step 6 — Convert to GB: 307,200,000 ÷ 1,073,741,824 ≈ 0.286 GB — since this is under 1 GB, the headline storage figure displays in MB instead.
Step 7 — Compare to a pricier model: the same 15,000,000 tokens on text-embedding-3-large ($0.13/1M) would cost 15 × $0.13 = $1.95 — about 6.5× more for the same corpus.
Total Embedding Cost
$0.3000
Cost / 1,000 Chunks
<$0.01
Vector Storage
293.0 MB

Explanation: At this volume, embedding cost is trivial ($0.30 total) because text-embedding-3-small is priced at just $0.02 per 1M tokens — the real budgeting question is storage, not API spend, since 293 MB of raw vectors will need to live in a vector database indefinitely and grow every time the knowledge base is updated. Switching to text-embedding-3-large would push the one-time embedding cost to $1.95 and, at its native 3,072 dimensions, roughly double the storage footprint to around 586 MB — a tradeoff worth making only if retrieval quality genuinely improves for your use case.

Interpretation

Understanding Your Embedding Cost Result

What your total embedding cost generally implies

Total Embedding CostWhat It Generally MeansRecommended Next Step
Under $10Small corpus or prototype-scale knowledge baseUse any model, including a premium one — cost isn't the constraint
$10 – $100Small production knowledge base or internal search toolProceed as planned; recheck cost if the corpus grows quickly
$100 – $1,000Meaningful corpus, likely tens of millions of tokensCompare model and dimension tradeoffs before committing
$1,000 – $10,000Large enterprise corpus or frequent re-embedding cadenceEvaluate self-hosted open-source models and negotiate volume pricing
Over $10,000Very large corpus, likely billions of tokensStrongly consider self-hosted embedding infrastructure to remove per-token fees

If your embedding cost looks tiny but storage looks large: that's normal and common — embedding is usually cheap per token, while the resulting vectors persist in your vector database indefinitely. Budget storage as an ongoing infrastructure line item, not a one-time cost like the embedding API charge.

If you're comparing models at very different price points: weigh the price gap against your retrieval quality requirements. A model 6x more expensive is only worth it if it measurably improves search relevance for your specific documents and queries — test on a sample before committing your whole corpus.

These are cost and storage estimates based on published per-token rates and raw float32 vector size, not exact billing or database figures. Always confirm current pricing and your vector database's actual per-vector overhead before finalizing a budget.

ℹ️

This calculator provides planning estimates only. Actual embedding charges depend on your provider's live pricing and exact tokenizer, and actual vector database storage costs depend on your provider's index overhead, replication, and metadata fields, which are not modeled here. Always verify against your provider's billing dashboard before finalizing a budget.

Use Cases

Practical Use Cases for the Embedding Cost Calculator

Where estimating embedding and vector storage cost up front genuinely helps

📚

Sizing a knowledge-base RAG system

Project the one-time embedding bill and ongoing storage footprint before building a support-docs or internal-wiki search feature.

🔁

Re-embedding cost after a model upgrade

Estimate what re-processing your entire corpus will cost when a provider ships a newer or higher-quality embedding model.

✂️

Choosing chunk size to control vector DB cost

Compare how smaller chunks with overlap increase total chunk count, embedding cost, and storage versus larger, non-overlapping chunks.

⚖️

Comparing embedding model providers

Run the same corpus across OpenAI, Cohere, and Google pricing to see which provider fits your budget and dimension needs.

💬

Budgeting a support chatbot's ingestion cost

Estimate the data-ingestion side of a RAG chatbot separately from the per-query generation cost covered by an LLM cost tool.

📈

Evaluating a dimension upgrade

Check whether moving from 768 to 1536 or 3072 dimensions is worth the extra storage and, for some providers, extra API cost.

🖥️

Planning GPU infra for self-hosted embeddings

See the $0 API-cost baseline for open-source models like BGE or E5, then budget GPU time separately for high-volume corpora.

💾

Estimating vector database storage bills

Use the MB/GB output as a starting point for your vector database's storage-based pricing tier or self-hosted disk budget.

🪆

Testing dimension truncation savings

Estimate how much storage a Matryoshka-style dimension reduction, such as 3072 down to 1024, would save on an existing corpus.

🏢

Multi-tenant SaaS embedding budgets

Project per-customer or per-workspace embedding cost for a SaaS product that embeds each tenant's documents separately.

🎓

Academic and research corpus indexing

Budget the cost of embedding a paper corpus or dataset for a research project's semantic search or clustering pipeline.

🗂️

E-discovery and compliance re-indexing

Estimate the cost of re-embedding a document set after a data classification change or legal e-discovery indexing requirement.

Pros & Cons

Benefits and Limitations

What this embedding cost calculator does well, and where it can't replace live billing or database data

✅ Benefits

  • Free, instant, and requires no signup or account
  • Runs entirely in your browser — your corpus details are never uploaded to a server
  • Covers cost across OpenAI, Cohere, Google, and self-hosted open-source models in one view
  • Estimates vector storage size alongside embedding cost, not just one or the other
  • Normalized cost-per-1,000-chunks figure makes comparing different corpus sizes easy
  • Side-by-side model comparison list at the same corpus size
  • Storage-by-dimension chart shows the tradeoff of a higher-dimension model visually
  • Bar chart compares total cost across five models instantly
  • Works for any corpus size, from a few hundred chunks to tens of millions
  • Downloadable plain-text summary of your estimate
  • Fast-loading and mobile-friendly
  • Useful as a repeatable check whenever provider pricing changes

⚠️ Limitations

  • Doesn't account for chunk overlap automatically — you must include overlapping tokens in your average manually
  • Doesn't include vector database index overhead, metadata, or replication (typically +20-50% on raw storage)
  • Doesn't model quantized or compressed vector storage (int8, binary), only standard float32
  • Self-hosted open-source models show $0 API cost since GPU/CPU cost varies by setup
  • Doesn't account for provider volume discounts or enterprise contracts
  • Doesn't model embedding API batch size or rate limits, which affect job duration, not cost
  • Pricing snapshots can lag a provider's most recent rate change
  • Not a substitute for your provider's live billing dashboard or vector database's storage metrics
Reference

Embedding Model Comparison

Dimensions, price per 1M tokens, and use-case fit across well-known embedding providers

ModelDimensionsPrice / 1M TokensBest Use-Case Fit
OpenAI text-embedding-3-small1536 (truncatable)$0.02High-volume RAG where cost per token matters most
OpenAI text-embedding-3-large3072 (truncatable)$0.13Maximum retrieval quality, technical/code search
Cohere embed-v41024 / 1536 (Matryoshka)$0.10Multilingual and multimodal (text + image) RAG
Google gemini-embedding3072 (configurable)$0.15Google Cloud-native pipelines, multilingual corpora
Voyage AI voyage-31024~$0.06 (varies by tier)Domain-tuned retrieval (legal, code, finance)
Open-source BGE-large / E5 (self-hosted)1024$0 API / GPU infra costCost-sensitive, high-volume, or data-residency-restricted deployments

Common Mistakes and Expert Tips

❌ Common Mistakes

  • Estimating cost from chunk count alone, ignoring average tokens per chunk, which is the real billing unit
  • Forgetting to account for chunk overlap, which embeds part of the same text more than once
  • Assuming a model or dimension switch is incremental, when it actually requires re-embedding the whole corpus
  • Budgeting only the embedding API cost and forgetting the ongoing vector database storage bill
  • Picking the highest-dimension model by default instead of testing whether it measurably improves retrieval
  • Treating self-hosted open-source models as free while ignoring GPU or CPU infrastructure cost

💡 Expert Tips & Best Practices

  • Cross-check your average tokens per chunk with the Token Count Calculator before running this estimate
  • Pair this calculator with the Cloud Storage Cost Calculator to budget the ongoing vector database storage bill
  • Use dimension truncation (Matryoshka) on OpenAI text-embedding-3 models to shrink storage without a full re-embed
  • Batch embedding API calls up to the provider's max batch size to reduce request overhead on large jobs
  • Re-run this calculator whenever you change chunk size or overlap strategy, since both directly change total tokens
📝

Summary: This embedding cost calculator gives you an instant, free projection of what it costs to embed a RAG corpus across five common providers, plus the vector storage footprint the resulting embeddings will occupy, so you can budget the data-ingestion side of a retrieval system with real numbers instead of guesswork. Pair it with the AI Token Calculator and Context Window Calculator for a complete RAG cost picture, from ingestion through generation.

FAQ

Frequently Asked Questions

Common questions about embedding cost and vector storage estimates

How much does it cost to embed a document corpus?
Cost depends on the total tokens in your corpus (chunks × average tokens per chunk) and the embedding model's price per 1M tokens. For example, OpenAI's text-embedding-3-small at $0.02 per 1M tokens costs about $0.30 to embed a 50,000-chunk corpus averaging 300 tokens per chunk (15 million tokens total), while text-embedding-3-large at $0.13 per 1M costs roughly 6.5x more for the same corpus.
How is vector storage size calculated?
Each embedding vector stores one 32-bit (4-byte) floating point number per dimension. Storage size is approximately chunks × dimension × 4 bytes. A 1,536-dimension embedding for 50,000 chunks needs roughly 50,000 × 1,536 × 4 = 307,200,000 bytes, or about 293 MB, before any index overhead.
Does a higher embedding dimension mean better search quality?
Generally yes, up to a point — higher-dimension embeddings can capture more nuance, but the improvement flattens out and storage/compute cost keeps rising linearly with dimension. Many teams find 768 or 1536 dimensions is a good balance between retrieval quality and infrastructure cost for most RAG applications.
Are open-source embedding models actually free?
There's no per-token API charge for self-hosted open-source models like BGE or E5, but you still pay for the GPU or CPU infrastructure to run them, plus engineering time for deployment and maintenance. Whether that's cheaper than a hosted API depends heavily on your volume and whether you already have spare GPU capacity.
Do I need to re-embed my whole corpus if I switch models?
Yes — embeddings from different models (and often different dimensions) are not directly comparable or interchangeable in the same vector index. Switching embedding models means re-processing your entire corpus and rebuilding your vector store, which is worth factoring into the true cost of a model change.
What is the embedding cost formula this calculator uses?
Total tokens equal chunks multiplied by average tokens per chunk. Total embedding cost equals total tokens divided by 1,000,000, multiplied by the selected model's price per 1M tokens. Cost per 1,000 chunks uses the same rate applied to 1,000 × average tokens per chunk, which lets you compare corpora of different sizes on equal footing.
What chunk size should I use before running this calculator?
Most RAG pipelines use 200-500 token chunks with some overlap (10-20%) between consecutive chunks to preserve context across boundaries. Smaller chunks improve retrieval precision but increase the total chunk count, and therefore both embedding cost and vector storage, since overlap means the same source text gets embedded more than once.
Why does chunk overlap increase my embedding cost?
Overlap deliberately re-includes a slice of the previous chunk's text at the start of the next chunk, so overlapping tokens get embedded twice. A 20% overlap on 300-token chunks adds roughly 60 redundant tokens per chunk, which increases both the number of chunks needed to cover a document and the total tokens billed.
How much does re-embedding a corpus after a model upgrade cost?
Re-embedding costs the same as the original embedding run, priced at the new model's rate, because the entire corpus must be re-processed — embeddings from different models are not interchangeable. Enter your existing chunk count and average tokens per chunk with the new model selected to get a like-for-like re-embedding cost estimate.
Does vector storage cost include the vector database itself?
No — this calculator estimates only the raw bytes of the embedding vectors (chunks × dimension × 4 bytes). It does not include your vector database's index structures, metadata fields, replication, or hosting fees, which typically add 20-50% more on top of the raw vector size, and vary significantly by provider.
Which embedding model is cheapest for a large RAG corpus?
Among hosted APIs in this calculator, OpenAI's text-embedding-3-small at $0.02 per 1M tokens is the cheapest, followed by Cohere embed-v4 at $0.10 and Google's gemini-embedding at $0.15, with text-embedding-3-large at $0.13 offering higher dimensionality at a higher price. Self-hosted open-source models like BGE or E5 have no per-token fee but require GPU infrastructure.
How do I estimate the number of chunks in my corpus?
Divide your total corpus size in tokens by your target chunk size, accounting for overlap. For example, a 10-million-token corpus split into 300-token chunks with no overlap yields roughly 33,333 chunks; adding 20% overlap increases that closer to 40,000-42,000 chunks since more chunks are needed to cover the same text.
Is embedding cost a one-time expense?
It's typically a large one-time cost for the initial corpus, plus a smaller recurring cost as new or changed documents are added. If your corpus updates frequently, budget for incremental re-embedding of only the changed chunks rather than re-running the full corpus each time, which keeps ongoing costs proportional to your update volume.
Can I shrink embedding dimension to save storage without re-embedding?
Only if your model supports it natively. OpenAI's text-embedding-3 family supports Matryoshka-style dimension truncation, letting you shorten the returned vector (for example from 3072 to 1024 dimensions) while keeping most of its retrieval quality, without re-calling the API. Most other models, including Cohere and open-source BGE/E5, require a full re-embed to change dimension.
Do embedding APIs have rate or batch size limits I should plan around?
Yes — providers cap both requests per minute and the number of text items per batch call (commonly a few thousand for OpenAI and Cohere). For a large corpus, this calculator's total cost figure is accurate, but your actual embedding job may need to run in batches over hours or days rather than a single call, which is a scheduling constraint separate from the dollar cost.
How accurate are the price estimates in this calculator?
Prices are current as of June 2026 and reflect each provider's published per-token rate for the listed model. Providers periodically revise embedding prices or add new model versions, so always verify current rates on the provider's official pricing page before finalizing a budget for a large embedding run.
Learn More

Authoritative Resources on Embeddings & Vector Storage

Official documentation to complement this calculator — always verify live rates before finalizing a budget

Related Calculators

Explore other AI & tech tools