Embeddings, priced.
Embedding a 100M-token knowledge base costs $1.00 on the cheapest model here — the bill that hurts is re-embedding it every time you switch models, and the vector index you store afterwards. Enter your corpus, churn and query volume and get every major embedding API ranked by first-year cost, batch discounts included.
Index jobs (the first embed, monthly changes, full re-indexes) can run on a vendor's batch endpoint; queries are real-time and always pay the standard rate. Dimensions and chunk size don't change the embedding bill — they size the vector index you'll store.
- Amazon Titan Text Embeddings V2$0.01/1M batch text $4.64/yr
- voyage-4-lite$0.0134/1M batch text $5.73/yr
- OpenAI text-embedding-3-small$0.02/1M text $7.84/yr
- voyage-4$0.0402/1M batch text $17.18/yr
- jina-embeddings-v5-text$0.05/1M text $19.60/yr
- Alibaba text-embedding-v4$0.07/1M text $27.44/yr
- Baichuan Text Embedding$0.0744/1M text $29.16/yr
- Amazon Nova Multimodal Embeddings$0.0675/1M batch multimodal $31.32/yr
List prices verified on vendor pages; oldest row 2026-08-31. Text-token rates only — image, audio and video embedding bill separately. Free allowances, prepaid-pack minimums and vector-database costs not modeled.
Six corpus profiles
Click one to load it. Corpus sizes, churn and query volumes are scenario assumptions you can edit; the verdicts are computed from the current rate sheet.
Your corpus against every model
The scenario configured above, priced on all 15 models. Rows keep the default-scenario order; rank and cost columns recompute live as you edit inputs.
| # | Model | Type | $ / 1M | Batch $ / 1M | Initial index | Per month | First year | Src |
|---|---|---|---|---|---|---|---|---|
| 1 | Amazon Titan Text Embeddings V2 Bedrock on-demand, US East. | text | $0.02 | $0.01 | $1.00 (batch) | $0.303 | $4.64 | ↗ |
| 2 | voyage-4-lite lite First 200M tokens free per account (one-time). Not in the math. | text | $0.02 | $0.0134 | $1.34 (batch) | $0.366 | $5.73 | ↗ |
| 3 | OpenAI text-embedding-3-small | text | $0.02 | — | $2.00 | $0.487 | $7.84 | ↗ |
| 4 | voyage-4 First 200M tokens free per account (one-time). Not in the math. | text | $0.06 | $0.0402 | $4.02 (batch) | $1.10 | $17.18 | ↗ |
| 5 | jina-embeddings-v5-text Prepaid packs: $50 per 1B tokens; $500 buys 11B. 10M-token starter key, non-commercial use only. Not in the math. | text | $0.05 | — | $5.00 | $1.22 | $19.60 | ↗ |
| 6 | Alibaba text-embedding-v4 1M free tokens, valid 90 days from activation. Not in the math. | text | $0.07 | — | $7.00 | $1.70 | $27.44 | ↗ |
| 7 | Baichuan Text Embedding Priced in CNY; converted at the catalog's dated rate. | text | $0.0744 | — | $7.44 | $1.81 | $29.16 | ↗ |
| 8 | Amazon Nova Multimodal Embeddings Text rate shown; images, audio and video bill separately. Bedrock, US East. | multimodal | $0.135 | $0.0675 | $6.75 (batch) | $2.05 | $31.32 | ↗ |
| 9 | voyage-4-large First 200M tokens free per account (one-time). Not in the math. | text | $0.12 | $0.0804 | $8.04 (batch) | $2.19 | $34.37 | ↗ |
| 10 | voyage-code-4 code First 200M tokens free per account (one-time). Not in the math. | text | $0.12 | $0.0804 | $8.04 (batch) | $2.19 | $34.37 | ↗ |
| 11 | Mistral Codestral Embed code | text | $0.15 | $0.075 | $7.50 (batch) | $2.27 | $34.80 | ↗ |
| 12 | Tencent Hunyuan Embedding Priced in CNY; converted at the catalog's dated rate. | text | $0.1042 | — | $10.42 | $2.54 | $40.85 | ↗ |
| 13 | Google Gemini Embedding 2 Text rate shown; images, audio and video bill separately. Free tier on the Gemini API is rate-limited, not a token quota. Not in the math. | multimodal | $0.20 | $0.10 | $10.00 (batch) | $3.03 | $46.40 | ↗ |
| 14 | ByteDance Skylark Embedding Vision Text rate shown; image input bills $0.325 per 1M tokens. | multimodal | $0.125 | — | $12.50 | $3.04 | $49.00 | ↗ |
| 15 | OpenAI text-embedding-3-large | text | $0.13 | — | $13.00 | $3.16 | $50.96 | ↗ |
"$ / 1M" is the text-token list price on the standard endpoint; "Batch" is the vendor's published batch rate, applied to index jobs when the switch is on. OpenAI, Tencent and Baichuan rows link to their pages in our model catalog, where the same rates are tracked. Src links open the vendor pricing page each rate was verified on.
The math behind the ranking
Every rate is a pay-as-you-go list price read off the vendor's own pricing page and dated. Rates for models already in our LLM catalog are read from it, so a reprice there moves this page too. The engine splits the bill into the four things that actually cost money:
corpus = documents × tokens_per_doc index_rate = batch_rate if batch is on and published, else list_rate initial = corpus × index_rate ÷ 1M updates/mo = corpus × changed% × index_rate ÷ 1M queries/mo = queries × tokens_per_query × list_rate ÷ 1M /* never batch */ reindex/mo = reindexes_per_year × corpus × index_rate ÷ 1M ÷ 12 first_year = initial + 12 × (updates + queries + reindex) /* model-independent */ vectors = corpus ÷ tokens_per_chunk index_bytes = vectors × dimensions × 4 /* float32 */
Only text tokens are compared; multimodal models bill images, audio and video at separate rates, listed in the table notes. Not modeled: free allowances (one-time, expiring or rate-limited), prepaid-pack minimums, chunk overlap, tokenizer differences between vendors, and the vector database itself. Amazon rates are the US East on-demand list, read from the AWS Price List API; Alibaba's are its International list.
Full methodology lives at /methodology/. Found a stale rate? Report it.
Three corpora, real numbers
Computed from the current rate sheet — recalculated on every data refresh.
What the per-token rate doesn't tell you
At list prices a 1B-token corpus indexes for $10.00–$130. The catch is that vectors from different models live in different spaces: changing models, or even chunk size, means re-embedding everything. The "re-indexes per year" field prices that in.
8 of 15 models publish a batch rate — Google, Amazon and Mistral at half price, Voyage at 33% off. Indexing is offline work, so it can wait for a batch window; a search query can't. OpenAI's pricing page lists no batch rate for its embedding models, so none is applied here.
Every model bills the tokens you send; none bills by output dimensions. But each vector is stored at its dimension count — 4 bytes per float32 — so a 3,072-dimension index is three times a 1,024-dimension one in your vector database. Some models let you request fewer — OpenAI's text-embedding-3 takes a dimensions parameter — and the dimensions field shows what that does to the index.
Voyage gives each account its first 200M tokens free, once. Alibaba's free tokens expire 90 days after activation. Gemini's free tier is rate-limited rather than a token quota, and Jina's starter key is for non-commercial use. None of them is a monthly allowance, so none enters the math.
Cohere now sells Embed 5 only as dedicated Model Vault instances, from $3/hour or $2,000/month — a capacity price that doesn't compare per token, so it isn't ranked. Jina sells prepaid packs ($50 per 1B tokens), which this calculator treats as a per-token rate.
Embedding pricing — answered.
The questions this calculator exists for, answered with the current numbers. Every figure recomputes when a vendor reprices.
Q · 01 How much do OpenAI embeddings cost? +
Q · 02 What is the cheapest embedding API? +
Q · 03 How much does it cost to embed 1 million documents? +
Q · 04 How much does Voyage AI cost? +
Q · 05 Is Gemini Embedding 2 free? +
Q · 06 How much storage does a vector index need? +
Q · 07 How much does Cohere Embed cost? +
Q · 08 Do I have to re-embed everything when I switch embedding models? +
Put this calculator on your site
Free to embed — your readers always see current rates. The widget links back here for the full rate table.
The cost toolbox
Every tool runs on the same live-pricing backbone. See the full index →