Skip to main content
An embedding model converts content into a fixed-length vector of numbers that represents its meaning. Content with similar meaning produces similar vectors, which is what allows a vector database to retrieve content by semantic similarity instead of keyword matching. Most embedding models take text as input. Multimodal embedding models also accept images, which they represent as vectors in the same space as text. Embedding models generate no language. They are used to build and query vector databases, and they cannot serve prompts or act as an agent’s cognitive engine. For models that generate text, see Generative models. Every embedding model in the catalog is available for inference and none can be fine-tuned. ibm-granite/granite-embedding-311m-multilingual-r2 is the default. Create a vector database without specifying a model and SeekrFlow uses the default.
The embedding model is fixed when you create a vector database. Every vector in a database is stored in that model’s vector space, and dimensions differ across models. An existing database cannot be switched to a different embedding model. To change models, create a new database and re-embed your documents.

​
Supported embedding models

Mixed means a model has no single country of origin. Empty cells are values the model catalog does not currently report. See Country of origin and Model attributes.
Bedrock embedding models are available for self-hosted AWS/EKS deployments only, and they are not listed in the SeekrFlow model catalog. See Use AWS Bedrock for ingestion and inference for setup instructions.
Models are added and updated over time. To retrieve the current models and their attributes, see List models.

​
Select an embedding model

Weigh the following when choosing a model. Because the choice cannot be reversed without re-embedding, settle it before you create the database.
  • Dimensions – Larger vectors capture more nuance and cost more to store and search. This value also defines the database’s vector space.
  • Max input tokens – Chunks longer than the model’s limit are not fully represented. Keep your chunk size within the limit. Avoid inputs longer than a model’s maximum input tokens.
  • Language support – Match the model to the languages in your corpus. An English-only model degrades on multilingual content.
  • Country of origin – Required if your deployment restricts model provenance. See Country of origin.

​
Create a vector database

To choose an embedding model and create a vector database:
Last modified on August 21, 2026