What Do Embeddings Cost?

Embedding models convert text into numeric vectors for semantic search, RAG pipelines, recommendations, and clustering — and they're billed on input tokens only, at rates from $0.02 to $0.15 per million tokens as of July 2026. That makes embeddings dramatically cheaper than LLM calls: embedding a 100,000-document knowledge base costs a few dollars, while answering questions over it with a flagship LLM can cost that much per day. This calculator converts your corpus size into a concrete indexing budget across every major provider.

Current Embedding Pricing (July 2026)

• OpenAI text-embedding-3-small: $0.02/M tokens (batch: $0.01) — the default choice for most projects
• OpenAI text-embedding-3-large: $0.13/M (batch: $0.065) — higher-dimension precision
• Voyage voyage-4-lite / 4 / 4-large: $0.02 / $0.06 / $0.12 per M (33% off via Files API)
• Cohere embed-v4: $0.12/M — multilingual strength, 128K-token context, no batch discount
• Google gemini-embedding-001: $0.15/M (batch: $0.075) — top MTEB multilingual scores
• Jina jina-embeddings-v3: $0.02/M — the budget alternative with solid benchmarks

How the Cost Is Calculated

English text averages roughly 1.33 tokens per word (equivalently, ~0.75 words per token) — the rule of thumb this calculator uses, and the same conversion behind our tokens-to-words converter. If you chunk documents before embedding — standard RAG practice, typically 300–800 tokens per chunk — count chunks as your documents here, since each chunk is a separate embedding call. Verify live rates on OpenAI's pricing page or your provider's equivalent before locking a budget.

How to Use This Calculator

Enter your document (or chunk) count and average words per document, pick a model, and you'll get the one-time indexing cost, total tokens, and cost per 1,000 documents. Tick the batch option if your indexing can run asynchronously — OpenAI and Google halve the price, Voyage takes a third off. Tick monthly re-indexing to see the annual figure: re-embedding frequency, not the one-time cost, is what turns a $30 index into a $360/year line item.

Worked Example

A RAG pipeline splits 100,000 support articles into 400-word chunks (one chunk per article on average). That's 100,000 × 532 tokens = 53.2M tokens. On text-embedding-3-small: 53.2 × $0.02 = $1.06 — effectively free. On gemini-embedding-001: $7.98. Even the premium tier barely registers, which is the real lesson: at typical corpus sizes, choose your embedding model on retrieval quality, not price. The budget conversation only starts at billions of tokens or high-frequency re-indexing.

The Costs This Calculator Doesn't Show

• Query-time embedding: every user search embeds the query text — tiny per call, but it's a perpetual cost that scales with traffic
• Vector storage: hosted vector databases commonly run $100–300/month at the million-vector scale — often 100× your embedding spend; higher-dimension models (3-large's 3072 dims) double or quadruple storage versus truncated alternatives
• Re-embedding on model migration: switching providers means re-indexing everything — vendor lock-in in embeddings is real, so choose deliberately
• The LLM on top: in a RAG system, generation dwarfs embedding costs — size that side with our LLM API cost calculator

Planning Your Full AI Budget

Embeddings are the cheap layer of an AI stack. Pair this estimate with the LLM API cost calculator for generation, the image generation cost calculator if visuals are involved, and the response length estimator to cap output spend. Building AI into a paid product? Feed the per-customer total into the SaaS pricing calculator so the costs land inside your margin instead of underneath it.

Frequently Asked Questions