FastEmbed

What it does

FastEmbed is a Python library that generates vector embeddings from text, images and document page screenshots. Inputs are strings, image file paths or PDF page images; outputs are numeric vectors, including dense text embeddings, sparse text embeddings, multivector (ColBERT-style) embeddings and relevance scores from reranking cross-encoders. Output vectors can be stored and queried in Qdrant with the qdrant-client package.

Developers use FastEmbed to embed a document collection, create a Qdrant collection sized to the model's output, upload vectors with payloads and run similarity queries. Unlike PyTorch-based toolkits, it runs on the ONNX Runtime, which keeps dependencies small enough for serverless environments such as AWS Lambda. GPU execution is available through a separate fastembed-gpu package that requires CUDA 12.x.

FastEmbed is aimed at developers building semantic search, retrieval-augmented generation or multimodal retrieval. It is installed with pip and runs locally in Python. Model choice matters: select a model from the supported-models list, or register a custom model with explicit configuration before embedding.

What it can do

• Generate dense text embeddings with supported models such as BAAI/bge-small-en-v1.5
• Produce sparse text embeddings, for example SPLADE++ via SparseTextEmbedding
• Create late-interaction multivector embeddings (ColBERT-style) with LateInteractionTextEmbedding
• Generate image embeddings and ColPali multimodal late-interaction embeddings
• Re-rank candidate documents with TextCrossEncoder models
• Register custom embedding or cross-encoder models with add_custom_model

Works with

• Qdrant (via qdrant-client[fastembed] and qdrant-client[fastembed-gpu])

Example workflows

Semantic search over documents with Qdrant

Input: a list of text documents and a model name. Steps: initialize QdrantClient, create a collection sized with get_embedding_size, upload documents as models.Document objects, then call query_points with a query document. Output: ranked matching points from the collection.

Rerank top retrieval results

Input: a query and a list of candidate documents. Steps: load TextCrossEncoder with a supported model such as Xenova/ms-marco-MiniLM-L-6-v2, call rerank with the query and documents. Output: a relevance score per document.

Before you choose

• GPU execution needs the separate fastembed-gpu package and CUDA 12.x; the docs point to an example page for troubleshooting.
• Model quality and dimensions depend on the chosen model, so compare supported models before indexing.

Listed under

• Coding
• Sandboxes and infrastructure
• Developers
• Embeddings
• Onnx
• Open source
• Python
• Python library

Get started

Run pip install fastembed, or pip install fastembed-gpu for GPU support.

1. Install the library with pip install fastembed, or pip install fastembed-gpu for GPU use.
2. For Qdrant integration, install pip install qdrant-client[fastembed] (or [fastembed-gpu]); quote the bracketed name on zsh.
3. Choose a supported model name (e.g. BAAI/bge-small-en-v1.5) or register a custom model with add_custom_model before running embed.

The software can be installed under its repository license. Infrastructure and model-provider charges remain separate. Optional commercial hosting and support have their own terms.

FastEmbed is free to use. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• FastEmbed - Qdrant ↗qdrant.tech
• GitHub - qdrant/fastembed: Fast, Accurate, Lightweight Python library to make State of the Art Embedding · GitHub ↗github.com

Company website

qdrant.tech: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.