Introducing zerank-1, our best reranker yet

We’re excited to announce the launch of zerank-1, our new state-of-the-art reranker that boosts top-k precision over any first-stage search, and outperforms models twice its size.

July 09, 2025 • Estimated Reading Time: 3 minutes

zerank-1 and zerank-1-small are live!

Rerankers play a critical role in surfacing the right documents from your top-k search results - reducing noise, latency, and LLM token usage. They refine search results (from a BM25, vector, or hybrid search) by re-scoring and reordering them based on query-document relevance. You can learn more about rerankers in our latest blog post here.

Today, we’re excited to release not one, but two rerankers: zerank-1 (4B parameters) and zerank-1-small (1.7B parameters). We trained these models using a brand new chess ELO score inspired pipeline, you can learn more about it here.

On benchmarks, zerank-1 and zerank-1-small outperform any first-stage search method across domains. They are also half as expensive as the leading closed source models out there like Cohere rerank-3.5 and Salesforce LlamaRank 8B.

zerank-1 and zerank-1 small improve the results of a hybrid (vector + keyword) search.

zerank-1 is now LIVE:

- Through our API (read the docs here)

- On HuggingFace

zerank-1-small is also available in Baseten’s library.

👉️ zerank-1 even performs better than prompting larger models like Gemini Flash 2.5 - delivering higher quality results with significantly lower latency and cost.

Our pricing is simple and transparent: $0.025 per million tokens.

👉️ 3 things we'd recommend checking out:

ZeroEntropy Team

At ZeroEntropy, we’re a team of researchers, mathematicians, and engineers focused entirely on solving AI search. Backed by Y Combinator and $4.2M in funding from Initialized Capital, we’re focused entirely on making search radically more accurate, efficient, and accessible.