## Monday, March 2, 2026

# Zembed-1: The Current Best Embedding Model

Umida Muratbekova

RAGEmbeddings

ZeroEntropy just released [zembed-1](/content/embeddings/zembed-1/index.html). It's a 4 billion parameter embedding model distilled from their zerank-2 reranker. We ran it through our evaluation suite to see how it performs on real retrieval tasks.

## TL;DR

[zembed-1](/content/embeddings/zembed-1/index.html) is now the **highest-ranked model** on our leaderboard, edging out [Voyage 4](/content/embeddings/voyage-4/index.html), [OpenAI text-embedding-3-large](/content/embeddings/openai-text-embedding-3-large/index.html), and [Jina Embeddings v5 Text Small](/content/embeddings/jina-embeddings-v5-text-small/index.html) in pairwise comparisons. **Strongest on general document retrieval and multilingual tasks.** Weaker on scientific and entity-heavy tasks.

## What We Found

### Best passage retrieval score

On MSMARCO - the standard information retrieval benchmark and the closest proxy to real RAG workloads - [zembed-1](/content/embeddings/zembed-1/index.html) hit 0.946 NDCG@10. That's the highest score across all 16 models in our leaderboard.

MSMARCO tests short queries against large document collections. A 0.946 means [zembed-1](/content/embeddings/zembed-1/index.html) consistently puts the right documents at the top.

### Wins most head-to-head matchups

We ran pairwise comparisons across all 15 other models. [zembed-1](/content/embeddings/zembed-1/index.html) wins 55-80% of matchups against most of the field. Against [Gemini text-embedding-004](/content/embeddings/gemini-text-embedding-004/index.html), it wins 80%. Against [Jina Embeddings v3](/content/embeddings/jina-embeddings-v3/index.html) and [Cohere Embed v3](/content/embeddings/cohere-embed-v3/index.html), around 67%.

The exceptions are the top 3: [Voyage 4](/content/embeddings/voyage-4/index.html), [OpenAI text-embedding-3-large](/content/embeddings/openai-text-embedding-3-large/index.html), and [Jina Embeddings v5 Text Small](/content/embeddings/jina-embeddings-v5-text-small/index.html). Against these, [zembed-1](/content/embeddings/zembed-1/index.html) is essentially even - win rates hover around 45-53%. No clear winner, which puts [zembed-1](/content/embeddings/zembed-1/index.html) in the same competitive tier as models that have been around longer.

### Performance varies by domain

[zembed-1](/content/embeddings/zembed-1/index.html) is not uniformly strong. Broken out by dataset:

The gap between general retrieval and scientific content is real. If your corpus is mostly research papers or knowledge-graph-style entity lookups, the top-tier models on those specific workloads still outperform [zembed-1](/content/embeddings/zembed-1/index.html) by a meaningful margin.

### How We Tested

- Each model embedded the same corpus and queries
- Retrieved top-10 results per query, calculated NDCG@10 and Recall@10
- Elo ratings from pairwise comparisons using GPT-5 as judge
- 5 datasets with human-annotated relevance: MSMARCO (general-purpose), FiQA (finance), SciFact (science), DBPedia (entities), ARCD (Arabic) + two private datasets

### Conclusion

[zembed-1](/content/embeddings/zembed-1/index.html) is the **strongest embedding model we've tested** on general-purpose retrieval, and it competes with the top tier across most general workloads. For RAG pipelines that handle unstructured documents, business content, or multilingual queries, it's worth testing.

We've added [zembed-1](/content/embeddings/zembed-1/index.html) to our [Embedding Leaderboard](/content/embeddings/index.html) where you can compare it against [Voyage 4](/content/embeddings/voyage-4/index.html), [OpenAI text-embedding-3-large](/content/embeddings/openai-text-embedding-3-large/index.html), [Cohere Embed v3](/content/embeddings/cohere-embed-v3/index.html), and others across all datasets.