LLM Calculators

Embeddings Cost Calculator

Compare embedding model pricing across OpenAI, Voyage AI, and Cohere.

The cost of creating vector embeddings depends on the model and your total token volume. OpenAI text-embedding-3-small costs $0.020 per million tokens — the cheapest OpenAI option. OpenAI text-embedding-3-large costs $0.130 per million tokens and delivers higher retrieval accuracy for semantic search. Voyage AI's voyage-3-lite starts at $0.010 per million tokens, making it the lowest-cost option for high-volume pipelines. Enter your corpus size below to calculate one-time and recurring embedding costs across all models — no signup required.

10,000
500

Total tokens to embed: 5,000,000

ModelProviderPrice/1MOne-Time CostDimensions
text-embedding-3-smallOpenAI$0.0200$0.10001,536
text-embedding-3-largeOpenAI$0.1300$0.65003,072
Cohere embed-v3Cohere$0.1000$0.50001,024
Voyage 3Voyage AI$0.0600$0.30001,024

How embedding costs are calculated

Embedding cost is determined by two variables: the total number of tokens in your corpus and the per-token price of the embedding model. Unlike LLM generation costs — which have separate input and output rates — embedding models charge a single flat rate per token processed. There is no output price because embeddings do not generate text; they convert text into a fixed-length vector. To estimate your total tokens, use the ratio of approximately 1 token per 4 characters of English text (or 250,000 tokens per 1MB of plain text). A corpus of 100,000 average documents — each 500 words / 3,750 characters — contains roughly 94 million tokens. At $0.020 per million tokens, that costs $1.88 to embed with text-embedding-3-small. The same corpus costs $12.22 with text-embedding-3-large. If you re-embed your corpus — for example, to update it with new content or switch to a newer model — that is an additional full-cost pass. Build re-embedding frequency into your budget: weekly re-embedding of a 100M-token corpus costs $2 per run with text-embedding-3-small, $13 per run with text-embedding-3-large.

Choosing the right embedding model

The three practical choices for most production RAG pipelines are OpenAI text-embedding-3-small, Voyage AI voyage-3, and Cohere embed-v4.0. Here is how they compare on the dimensions that matter: **OpenAI text-embedding-3-small** ($0.020/M tokens) is the default choice for teams already on the OpenAI platform. It is performant on standard retrieval benchmarks, requires no additional vendor relationship, and is priced aggressively for its quality tier. The main limitation is that it produces 1536-dimensional vectors — larger than some alternatives, which increases vector storage costs at scale. **Voyage AI voyage-3-lite** ($0.010/M tokens) is the cheapest option for pure throughput. It trades some retrieval accuracy for a 50% cost reduction versus text-embedding-3-small. Best fit: high-volume pipelines where retrieval precision is less critical than cost, such as broad content discovery or deduplication. **Voyage AI voyage-3** ($0.060/M tokens) outperforms text-embedding-3-small on most domain-specific benchmarks — particularly code, legal, and scientific text. If your RAG quality matters and you are not locked into OpenAI infrastructure, voyage-3 is the best quality-per-dollar option in most evaluations. **OpenAI text-embedding-3-large** ($0.130/M tokens) delivers the highest retrieval accuracy of the mainstream options but at 6.5× the cost of text-embedding-3-small. Justified only for applications where retrieval precision directly drives revenue — such as enterprise search, legal discovery, or high-stakes recommendation engines.

Frequently Asked Questions

How much does OpenAI text-embedding-3-small cost per million tokens?
OpenAI text-embedding-3-small costs $0.020 per million tokens. This is the most cost-efficient OpenAI embedding model. By comparison, text-embedding-3-large costs $0.130 per million tokens — 6.5× more expensive — and the older ada-002 model costs $0.100 per million tokens. For most RAG pipelines, text-embedding-3-small delivers sufficient retrieval quality at a fraction of the cost of the larger model.
What is the cheapest embedding model?
Voyage AI voyage-3-lite is the lowest-cost mainstream embedding model at $0.010 per million tokens — half the price of OpenAI text-embedding-3-small ($0.020/M). For high-volume embedding pipelines where cost is the primary constraint, voyage-3-lite is the best starting point. OpenAI text-embedding-3-small is the cheapest option if you want to stay within the OpenAI ecosystem.
How many tokens are in 1MB of plain text?
One megabyte of plain English text contains approximately 250,000 tokens. This is based on the standard approximation of 1 token per 4 characters of English. For reference: a 500-word document is approximately 3,750 characters, or about 938 tokens. A corpus of 100,000 such documents totals roughly 94 million tokens. Code and CJK languages tokenize differently — code is typically denser (1 token per 3–3.5 characters), while Chinese, Japanese, and Korean text is sparser (1 token per 2–2.5 characters).
What is the difference between text-embedding-3-small and text-embedding-3-large?
text-embedding-3-small and text-embedding-3-large are both OpenAI's current embedding models, but they differ in output dimension, cost, and retrieval accuracy. text-embedding-3-small produces 1536-dimensional vectors and costs $0.020 per million tokens. text-embedding-3-large produces 3072-dimensional vectors and costs $0.130 per million tokens. The large model scores higher on MTEB benchmarks, particularly for multilingual retrieval and domain-specific tasks. For most general-purpose RAG applications, text-embedding-3-small achieves 85–90% of the large model's retrieval quality at 15% of the cost.
How much does it cost to embed 1 million documents?
The cost depends on average document length and the model chosen. For a corpus of 1 million documents averaging 500 words each (approximately 938 tokens per document), the total token volume is roughly 938 million tokens. At current pricing: text-embedding-3-small costs $18.76, text-embedding-3-large costs $121.94, voyage-3-lite costs $9.38, and voyage-3 costs $56.28. Use the calculator above to enter your specific document length and corpus size for an exact estimate.
Do I pay again if I re-embed my corpus?
Yes. Every embedding run — initial or re-run — charges the full per-token rate. There is no caching or re-use discount for previously embedded content. If you re-embed your entire corpus to switch models, update stale content, or add new documents, you pay for every token processed in that run. This makes model selection important upfront: switching embedding models at scale is not free. A 1-billion-token corpus re-embedded with text-embedding-3-small costs $20 per full re-run.
How does Voyage AI's pricing compare to OpenAI for embeddings?
Voyage AI's cheapest model (voyage-3-lite at $0.010/M tokens) is half the price of OpenAI text-embedding-3-small ($0.020/M tokens). Voyage's mid-tier model (voyage-3 at $0.060/M tokens) is between text-embedding-3-small and text-embedding-3-large on cost, but consistently outperforms text-embedding-3-small on domain-specific retrieval benchmarks including code, legal, and scientific text. For teams not locked into OpenAI infrastructure, Voyage AI typically offers better quality-per-dollar at the mid tier.

Explore Related Tools

Fine-Tuning Cost

Quick GPT and LLM fine-tuning training cost estimate from tokens and epochs.

Batch API Cost

Estimate OpenAI Batch API savings vs standard API at your request volume.

Cost Estimator

Estimate monthly and annual AI API costs based on your usage pattern.

All pricing data sourced from official provider pricing pages and verified manually. Prices subject to change — this calculator updates with each monthly pricing refresh. Token estimates use standard approximations: 1 token ≈ 4 characters (English), 1 token ≈ 3.5 characters (code), 1 token ≈ 2.5 characters (CJK). Verify exact token counts with provider tokenizers for production cost planning.