Which embedding model is cheapest for RAG?
Trying to find the cheapest embedding model for RAG? Picking the wrong model can multiply your bill for identical output. This comparison is for teams building retrieval and ranks the realistic options side by side using current prices, so the cheapest choice for your workload is obvious.
Results update automatically as you type.
| text-embedding-3-small | OpenAI | $0.02/M | 1,536 | $0.2000 |
| Embed v3 (English) | Cohere | $0.1/M | 1,024 | $1.00 |
| text-embedding-3-large | OpenAI | $0.13/M | 3,072 | $1.30 |
Why this isn't trivial
The part people underestimate: embedding cost scales with corpus size and dimensions, and higher dimensions also raise vector storage cost. In practice the biggest savings come from choosing a low-price, lower-dimension model that still retrieves well, so it is worth modelling before you commit.
How it's calculated
We estimate this by comparing per-token embedding rates and the storage implied by each model's dimensions. Every figure uses the current provider prices baked into the site (reviewed daily), and you can override any input to match your own assumptions.
Frequently asked questions
Do more dimensions mean better retrieval?+
Not always — many tasks do fine on smaller embeddings, saving on storage too.
What costs more, embedding or storage?+
Depends on corpus size and query volume; the comparison shows both.
Are these prices up to date?+
Yes. The model prices behind this calculator are refreshed and reviewed daily, so your estimate reflects current provider rates rather than a stale snapshot.
Related
Estimates for planning. Pricing data last reviewed 28 July 2026.