AI Potluck
Back to Gap Map Model components / Embeddings & retrieval models

zerank

ZeroEntropy
open weights / Overall score: 3.7

ZeroEntropy's reranker line. zerank-2 is a 4B Qwen3-derived cross-encoder with a 32K context, trained through a multi-stage pipeline that models query-document relevance as adjusted Elo ratings; its card reports an average 0.6714 NDCG@10 across seven domains against 0.5847 for Cohere rerank-3.5. The earlier zerank-1 and zerank-1-small ship alongside it. Notion acquired ZeroEntropy in July 2026 and the weights were released under Apache 2.0.

Added 2026-09-11 for the embeddings_retrieval promotion. Two claims were checked rather than taken on trust, and both hold - the zerank-2 card front matter and its licence section both read apache-2.0, and ZeroEntropy's own announcement records the Notion acquisition and says all models are now open source under Apache 2.0. Note the hosted ZeroEntropy API was sunset on 2026-09-04; the weights product is unaffected and stays on the Hub, so no end_of_life is declared. Verified 2026-09-11 via the zerank-2-reranker model card and the ZeroEntropy announcement.

Openness

3 high confidence
3.0
weights
open(Apache-2.0 ungated safetensors for zerank-1, zerank-1-small and zerank-2)
data
closed(the Elo-based relevance pipeline is described in arXiv 2509.12541
code
closed(no training or model repository
license
Apache-2.0(OSI)

Both claims that came with this product were checked rather than assumed, and both hold. The zerank-2 card states Apache-2.0 in its front matter, its Model Details table and a closing License section; ZeroEntropy own announcement records the Notion acquisition and says all its models are now fully open source under Apache 2.0. No corpus and no code, so an OSI licence over downloadable weights lands on the open-weights fallthrough.

Adoption

3 high confidence
3.0

About 657K downloads in the trailing 30 days across the line, almost all of it zerank-2 at 645K, with zerank-1-small at 15K and zerank-1 at 1.2K. Squarely inside the 100K-1M band on zerank-2 alone.

Capability

4 medium confidence
4.0

Rung 4, competitive frontier - a current 4B model beating the named proprietary rerankers on the publisher own seven-domain suite. Level with jina-reranker, which is the other 2025-26 frontier line here. Confidence is medium rather than high because the instrument is a vendor-constructed domain suite with no public leaderboard behind it, unlike BEIR or MMTEB.

  • https://huggingface.co/zeroentropy/zerank-2-reranker/raw/main/README.md recorded 2026-09-11

    Evaluations table gives zerank-2 an average NDCG@10 of 0.6714 across seven domains against 0.6456 for zerank-1, 0.5999 for Gemini 2.5 Flash listwise, 0.5847 for Cohere rerank-3.5 and 0.4509 for the OpenAI embedding retriever alone; 4B parameters, 32,768-token context.

Verified 2026-09-11