Bedrock models with embedding models
These Amazon Bedrock models turn text into vectors for semantic search, clustering and deduplication. They do not generate text — an embedding model is the retrieval half of a RAG system, and always needs a text model beside it to answer with.
10 of the models on Bedrock qualify today, from $0.100 per 1M input tokens. Rebuilt daily from the AWS catalogue, so this list does not go stale.
What teams get wrong about this
Embeddings are billed on input only, with no output cost, which makes them look almost free per call. The cost that surprises teams is re-indexing: changing embedding model means re-embedding the entire corpus, and vectors from two different models are not comparable. Choose deliberately, because switching later is a full rebuild.
All 10 models, cheapest first
| Model | Input /1M | Max input |
|---|---|---|
| Titan Embeddings G1 - Text amazon.titan-embed-text-v1 | $0.100 | 8K |
| Amazon Nova Multimodal Embeddings amazon.nova-2-multimodal-embeddings-v1:0 | — | — |
| Titan Multimodal Embeddings G1 amazon.titan-embed-image-v1 | — | — |
| Titan Text Embeddings v2 amazon.titan-embed-g1-text-02 | — | — |
| Titan Text Embeddings V2 amazon.titan-embed-text-v2:0 | — | 8K |
| Embed English cohere.embed-english-v3 | — | 1K |
| Embed Multilingual cohere.embed-multilingual-v3 | — | 1K |
| Embed v4 cohere.embed-v4:0 | — | 128K |
| Marengo Embed 3.0 twelvelabs.marengo-embed-3-0-v1:0 | — | — |
| Marengo Embed v2.7 twelvelabs.marengo-embed-2-7-v1:0 | — | — |
Capability flags come from the AWS Bedrock model catalogue; prices from the Price List API. Region counts include cross-region inference profiles. Methodology.
Questions
What is the cheapest embedding model on Bedrock?
The table is sorted cheapest-first on input price per 1M tokens. Note that embedding prices vary more by region than text models do — some are twice the price in Asia Pacific regions compared with us-east-1.
Can I mix embedding models in one index?
No. Vectors from different models live in different spaces and their distances are meaningless against each other. One index, one embedding model, and a model change means re-embedding everything.