zembed-1-embedding

by zeroentropy

Domain-specialist 4B embedding model distilled from reranker for finance, legal, healthcare, and code

98Kdl/month

4Bparams

HuggingFace Use in Pipeline

Identifiers

Model ID

zeroentropy/zembed-1-embedding

Feature URI

mixpeek://text_extractor@v1/zeroentropy_zembed_1_v1

Overview

zEmbed-1 is a 4B-parameter text embedding model built on Qwen3-4B and distilled from ZeroEntropy's zerank-2 reranker using an ELO-inspired training methodology. It leads on domain-specific benchmarks — outperforming Cohere Embed v4, OpenAI text-embedding-3-large, and Gemini Embedding on finance (0.4476), healthcare (0.6260), legal (0.6723), code (0.6452), and STEM (0.5283) retrieval tasks.

On Mixpeek, zEmbed-1 is the top choice for regulated-industry search where domain accuracy matters more than model size. Its flexible output dimensions (2560 down to 40) and support for binary quantization enable deployment from cloud to edge.

Architecture

Qwen3-4B backbone with task-specific encode_query() and encode_document() prompting. 4B parameters. 32K token context. Flexible projection head supporting 7 output dimensions (2560, 1280, 640, 320, 160, 80, 40). Trained via zELO methodology using adjusted Elo ratings for relevance scoring, distilled from zerank-2 cross-encoder.

Mixpeek SDK Integration

from mixpeek import Mixpeek

mixpeek = Mixpeek(api_key="YOUR_API_KEY")

mixpeek.ingest.documents(
    collection="sec_filings",
    source={"type": "s3", "bucket": "finance-docs"},
    pipeline={
        "embedding": {
            "model": "mixpeek://text_extractor@v1/zeroentropy_zembed_1_v1"
        }
    }
)

Capabilities

Best-in-class domain retrieval for finance, healthcare, legal, code, and STEM
32K token context length
7 flexible embedding dimensions (2560 down to 40)
50+ language support with >50% non-English training data
Binary quantization support for edge deployment

Use Cases on Mixpeek

Financial document search across SEC filings, earnings calls, and research reports

Clinical trial and medical literature retrieval

Legal case law search with nuanced regulatory understanding

Code search across enterprise repositories

Benchmarks

Dataset	Metric	Score	Source
Finance Retrieval	nDCG@10	0.4476	Model card
Healthcare Retrieval	nDCG@10	0.6260	Model card
Legal Retrieval	nDCG@10	0.6723	Model card
Code Retrieval	nDCG@10	0.6452	Model card