AI Rate CardEmbedding API pricing: what text embeddings cost at every provider
Updated Oct 17, 2026
An embedding model turns text into a list of numbers so a search engine, a RAG pipeline or a recommender can find similar passages. You pay only for the text you send in: there is no output price, and the vectors themselves are free. That makes embeddings the cheapest thing you can buy from an AI provider - and the prices still differ twenty-six times from the bottom of the market to the top. The tables below are rebuilt from today's prices on every page view.
The price in one line
In the 17 October 2026 collection the typical (median) embedding model cost $0.05 per 1M input tokens. The small open models (MiniLM, BGE base, E5 base) start at $0.005, and OpenAI's large model is $0.13.
What that means for a real job: indexing 10 million tokens (roughly 7,500 pages of text) costs about $0.10 with Qwen3 Embedding 8B, $0.20 with text-embedding-3-small and $1.30 with text-embedding-3-large. You index once and pay again only for new documents and for the queries, so for most projects embeddings are a rounding error next to the chat model.
OpenAI text-embedding-3-small: every provider
| Provider | Model | Input, $ | Output, $ |
|---|---|---|---|
| cometapi | Text Embedding 3 Small | $0.016 | $0.016 |
| openai | text-embedding-3-small | $0.02 | — |
| aihubmix | Text Embedding 3 Small | $0.02 | $0.02 |
| azure | text-embedding-3-small | $0.02 | — |
| fastrouter | OpenAI: text-embedding-3-small | $0.02 | — |
| impossibl | Text Embedding 3 Small | $0.02 | — |
| vercel | text-embedding-3-small | $0.02 | — |
| zenmux | OpenAI: Text Embedding 3 Small | $0.02 | — |
| trustedrouter | OpenAI Text Embedding 3 Small | $0.0211 | — |
Live data: this table is built from today's prices.
$0.02 at OpenAI, Azure, Vercel, FastRouter and most routers; CometAPI listed $0.016.
OpenAI text-embedding-3-large: every provider
| Provider | Model | Input, $ | Output, $ |
|---|---|---|---|
| cometapi | Text Embedding 3 Large | $0.10 | $0.10 |
| openai | text-embedding-3-large | $0.13 | — |
| aihubmix | Text Embedding 3 Large | $0.13 | $0.13 |
| azure | text-embedding-3-large | $0.13 | — |
| fastrouter | OpenAI: text-embedding-3-large | $0.13 | — |
| impossibl | Text Embedding 3 Large | $0.13 | — |
| vercel | text-embedding-3-large | $0.13 | — |
| zenmux | OpenAI: Text Embedding 3 Large | $0.13 | — |
| trustedrouter | OpenAI Text Embedding 3 Large | $0.14 | — |
Live data: this table is built from today's prices.
$0.13 almost everywhere, $0.104 at CometAPI. Six and a half times the small model: take it only if your own test shows better search results on your data.
Qwen3 Embedding 8B (open weights): every provider
| Provider | Model | Input, $ | Output, $ |
|---|---|---|---|
| deepinfra | Qwen3 Embedding 8B | $0.01 | — |
| fastrouter | Qwen: Qwen3 Embedding 8B | $0.01 | — |
| huggingface | Qwen 3 Embedding 8B | $0.01 | — |
| nebius | Qwen3-Embedding-8B | $0.01 | — |
| vercel | Qwen3 Embedding 8B | $0.01 | — |
| moark | Qwen3 Embedding 8B | $0.01049 | $0.01049 |
| trustedrouter | Qwen3 Embedding 8B | $0.01055 | — |
| tensorx | Qwen3 Embedding 8B | $0.02 | — |
| aihubmix | Qwen3 Embedding 8B | $0.068 | — |
| novita | Qwen3 Embedding 8B | $0.07 | — |
| infomaniak | Qwen3 Embedding 8B | $0.08537 | — |
| impossibl | Qwen3 Embedding 8B | $0.10 | — |
| ovhcloud | Qwen3 Embedding 8B | $0.12 | — |
| regolo-ai | Qwen3 Embedding 8B | $0.001118 / request | — |
Live data: this table is built from today's prices.
$0.01 at DeepInfra, FastRouter, Hugging Face, Nebius and Vercel - half of text-embedding-3-small, with a 32K-token context against 8K. Others ask up to $0.12 for the same model, so the provider matters more than the model here.
How to choose
- A first RAG or search prototype: text-embedding-3-small or Qwen3 Embedding 8B - both cheap, both good.
- Long documents: a model with a long context (Qwen3 Embedding 8B reads 32K tokens at once) needs fewer chunks.
- Millions of short texts (product titles, tickets): a small open model at $0.005 is often enough.
- Never mix models in one index: vectors from two models are not comparable - switching means re-indexing everything. To estimate a month, put your own volume into the cost calculator.
All prices are in USD per 1 million input tokens unless the row says otherwise, without taxes, as the providers publish them; each row links to its card with the source and the date it was checked.