Embeddings API pricing
Cost per 1M tokens for every major text-embedding API — OpenAI text-embedding-3, Gemini, Cohere Embed v4, Voyage, Mistral and open-weight BGE-M3. Plus output dimensions and max input length, the two specs that decide retrieval quality and storage cost.
Embedding API pricing — cost per 1M tokens
Cheapest first. Enter your corpus size to estimate the one-time embedding bill. All prices USD, last verified 2026-06-27.
| Model | Provider | Per 1M | Embed cost | Dims | Max input |
|---|---|---|---|---|---|
| BGE-M3open | BAAI | $0.010 | — | 1024 | 8K |
| text-embedding-3-small | OpenAI | $0.020 | — | 1536 | 8K |
| Jina Embeddings v3 | Jina AI | $0.020 | — | 1024 | 8K |
| Voyage-3.5 | Voyage AI | $0.060 | — | 1024 | 32K |
| Mistral Embed | Mistral | $0.100 | — | 1024 | 8K |
| Cohere Embed v4multimodal | Cohere | $0.120 | — | 1536 | 128K |
| text-embedding-3-large | OpenAI | $0.130 | — | 3072 | 8K |
| Gemini Embeddingmultimodal | $0.150 | — | 3072 | 2K |
Embedding cost = price × documents × tokens/doc. Cheapest row highlighted. Open-weight models can be self-hosted for compute cost only.
Pricing for each embedding model
BGE-M3 · BAAI
API string: BAAI/bge-m3 · also: BGE M3, bge-m3, BAAI embedding
Open weights — self-host for free; multilingual, dense + sparse + multi-vector. $0.010 per 1M tokens · 1024 dims · 8,192-token max input.
text-embedding-3-small · OpenAI
API string: text-embedding-3-small · also: OpenAI embedding small, text-embedding-3-small
Cheap default; Matryoshka — shorten dimensions to save storage. $0.020 per 1M tokens · 1536 dims · 8,191-token max input · $0.010 batch.
text-embedding-3-small pricing details → · Official pricing ↗
Jina Embeddings v3 · Jina AI
API string: jina-embeddings-v3 · also: Jina v3, jina-embeddings-v3
Multilingual with task-specific LoRA adapters. $0.020 per 1M tokens · 1024 dims · 8,192-token max input.
Voyage-3.5 · Voyage AI
API string: voyage-3.5 · also: Voyage 3.5, voyage-3.5, Voyage embedding
Top retrieval quality per dollar; 32K context. $0.060 per 1M tokens · 1024 dims · 32,000-token max input.
Mistral Embed · Mistral
API string: mistral-embed · also: mistral-embed, Mistral embedding
Solid multilingual embeddings from Mistral. $0.100 per 1M tokens · 1024 dims · 8,000-token max input.
Cohere Embed v4 · Cohere
API string: embed-v4.0 · also: Embed v4, embed-v4.0, Cohere embedding
Multimodal (text + image), long 128K context, Matryoshka dimensions. $0.120 per 1M tokens · 1536 dims · 128,000-token max input.
text-embedding-3-large · OpenAI
API string: text-embedding-3-large · also: OpenAI embedding large, text-embedding-3-large
Highest-quality OpenAI embedding; 3072 dims (shortenable). $0.130 per 1M tokens · 3072 dims · 8,191-token max input · $0.065 batch.
text-embedding-3-large pricing details → · Official pricing ↗
Gemini Embedding · Google
API string: gemini-embedding-001 · also: gemini-embedding-001, Gemini embedding, Google embedding
Multimodal embeddings; Matryoshka (768 / 1536 / 3072). $0.150 per 1M tokens · 3072 dims · 2,048-token max input.
How embedding pricing works
Embedding APIs charge per input token only — there is no output price, because the model returns a fixed-length numeric vector rather than generated text. Your bill is simply price per 1M × total tokens embedded. The big cost events are the initial corpus embedding and any re-embedding when you switch models (vectors are not portable between models).
Two specs drive the downstream cost as much as the per-token price: dimensions (higher = marginally better retrieval but proportionally more vector-DB storage and slower queries) and max input (how much text fits in one request before you must chunk). Several models support Matryoshka dimension shortening, letting you keep a high-quality model but store smaller vectors.
Embeddings API pricing FAQ
How much do text embeddings cost?
Embedding APIs bill per input token, typically $0.010–$0.150 per 1M tokens. Embedding 1 million ~500-token documents (500M tokens) costs roughly $5.00 on the cheapest model and $75.00 on the priciest.
What is the cheapest embedding API?
BGE-M3 (BAAI) is the cheapest tracked at $0.010 per 1M tokens. Open-weight models like BGE-M3 can be free if you self-host. OpenAI text-embedding-3-small ($0.02) is the cheapest first-party option.
Do more dimensions mean better embeddings?
Higher dimensions give marginally better retrieval but cost proportionally more to store and query in a vector database. 768–1024 dimensions are enough for most apps; OpenAI, Cohere and Gemini support Matryoshka shortening so you can pick a smaller size from a larger model.
Is embedding billed for input and output tokens?
Only input. Embeddings return a fixed-size numeric vector, not generated text, so there are no output-token charges — cost depends purely on how many tokens you send.