FastEmbed

Choose a model for the input and vector type you need. FastEmbed runs embedding models locally in your Python application, so you can send the resulting vectors to any Qdrant deployment.

These tables list the built-in embedding models in FastEmbed 0.9.0. Model names, dimensions, and licenses come from that release’s model registry. Your installed version determines which models you can use.

Dense Text Models

These models produce one dense vector per text input. Use the listed dimensions when you create a collection.

Follow Generate Text Embeddings to load a model, or Dense Embeddings with Qdrant to index and search its vectors.

Sparse Text Models

These models produce sparse vectors. Configure a sparse vector field in your collection.

See the miniCOIL and SPLADE guides for indexing and search examples.

Text Multivector Models

These models produce multiple vectors per text input. Dimensions refer to each vector in the output.

Follow ColBERT Multivector Embeddings to configure a collection and search with multivectors.

Image Models

These models produce one dense vector per image.

Image and Text Multivector Models

These models support image and text inputs and produce multivectors. Dimensions refer to each vector in the output.

Store and Search the Vectors

Next Steps

• Explore the FastEmbed guides for embedding generation and Qdrant integration.
• Rerank with FastEmbed to score retrieved results with a cross-encoder.
• Choose an inference approach to compare local embedding generation with Qdrant Cloud Inference.
• Filter search results by payload fields.
• Combine dense and sparse vectors for hybrid search.
• Reduce memory use with quantization.