Embedding & Reranker Models, Compared
Models that turn text into vectors for semantic search and RAG, plus rerankers that sort results by relevance. Each page lists downloads, licence and how to serve it.
✓ 0 live on LLM API · 70 listed · updated weekly
1. What are you building?
2. Where will it run?
3. What matters more?
How to choose an embedding model
1. Match the language
English-only models are smaller and often sharper on English text. If your documents mix languages, pick a multilingual model so queries and documents land in the same vector space.
2. Size versus speed
Small models embed millions of chunks cheaply on a CPU; larger ones retrieve better but need a GPU. Start small and move up only if search quality falls short.
3. Add a reranker for precision
Retrieve 50–100 candidates with an embedding model, then let a reranker score the top ones. It is the simplest way to lift RAG answer quality.
Frequently asked questions
What is the difference between an embedding model and a reranker?
An embedding model turns each text into a vector once, so you can search millions quickly. A reranker reads the query and a passage together and scores them — slower, but more precise on a short list.
Can I change embedding models later?
Yes, but vectors from different models are not compatible, so you have to re-embed your documents when you switch.
How are embedding models rated on this page?
No independent lab publishes comparable scores or prices for most of these models, so the rating uses one public number: Hugging Face downloads over the last 30 days. More downloads give a higher score on a 1.0–5.0 scale. About 10,000 downloads a month is roughly 2.4, 1 million is 3.6, 10 million is 4.2, and 100 million or more is close to 4.8. Hosted models without a Hugging Face download count keep the standard rating.
Can I run these myself?
Every model here has published weights. Each page lists the licence and the commands to serve it.
What does the rating for embeddings & rerankers models mean?
Each model is rated from 1.0 to 5.0 specifically for embeddings & rerankers. Green is 4.0 and above, yellow 3.0–3.9, red below 3.0. The same model can score differently in another category.
How is the rating calculated?
It comes from public data, not paid reviews: 50% production adoption (public leaderboard standing, routing popularity and download volume), 30% the benchmark that matters for this category, and 20% tooling and value for money (tool calling, structured outputs, context size and price against the category median).
What does the ✓ next to a rating mean?
It means an independent lab — LMArena or Artificial Analysis — has tested the model. Ratings without it are estimated from published specs such as context size, tooling support and price.
How are brand-new frontier models rated?
When a new model has strong independent benchmark results but no public usage data yet, adoption is left out instead of counted as low, and the rating uses 60% benchmark and 40% tooling and value.
How often are ratings updated?
Every week. If a source can't be reached, the previous week's figure is kept.
