Reranking with `zerank-2`
Use `zerank-2` for reranking retrieval candidates so the best matches reach the model. The homepage describes it as the flagship reranker and says it can improve retrieval with a single line of code.
ZeroEntropy
ClaimZeroEntropy builds specialized AI models for retrieval-heavy systems, including rerankers, embeddings, and custom models. It supports self-serve API use, enterprise licensing, and VPC deployment for production AI teams.
What ZeroEntropy is
ZeroEntropy is a retrieval-focused AI product that trains specialized models for search and RAG pipelines. Its core line includes rerankers, embeddings, and custom models designed for production systems where accuracy, latency, and cost all matter.
The homepage positions the stack as a faster and more accurate alternative to generalist models, while the pricing and trust pages show how teams can adopt it through self-serve APIs, enterprise plans, VPC deployment, and compliance documentation.
Core capabilities
Embedding models
Use `zembed-1` for embeddings when your system needs retrieval vectors rather than only reranking. The site positions it as the flagship embedding model and says it can outperform leading embedding models even at lower dimensionality.
Custom-trained models
Train specialized models for query rewriting, context compression, and bespoke production-agent workflows. The homepage frames custom models as a way to adapt the stack to a specific application rather than using a general-purpose model unchanged.
Multiple deployment paths
Access the models through a single API, partner providers, or dedicated VPC deployment. The pricing page also mentions AWS Marketplace and Azure as deployment channels for ZeroEntropy VPC.
Search infrastructure API
Use the search API for retrieval workflows that include OCR, indexing, storage, queries, reranking, and embeddings. The pricing page breaks these into usage-based components for teams that want retrieval infrastructure rather than one isolated model.
Benchmark evaluation tooling
Evaluate models against public benchmarks and internal datasets. The evaluations page shows rankings across 28 datasets and multiple verticals, which gives teams a way to compare retrieval quality before production rollout.
Common ways teams use ZeroEntropy
RAG and retrieval pipelines
Improve the ordering of retrieval candidates before they reach an LLM, especially when keyword and vector search both contribute documents and ranking quality determines answer quality.
Customer support automation
Power support systems where wrong retrieval creates bad responses. The Assembled story shows production reranking across chat, email, and phone support workflows.
Research search over large corpora
Support medical or other research-heavy search systems that must retrieve from large document collections and maintain accuracy under domain-specific queries. The site cites Vera Health using ZeroEntropy for retrieval across millions of medical research papers.
Low-latency agent workflows
Reduce latency in interactive AI products and agents by replacing slower generalist alternatives with specialized models optimized for serving performance.
Model evaluation and rollout
Run evaluation-driven model selection before production rollout. The evaluations page and Assembled case study both emphasize benchmark review, regression sets, and live traffic validation.
Pros and Cons
Pros
- Offers specialized rerankers, embeddings, and custom models in one product family.
- Supports several adoption paths, including self-serve API, partner providers, enterprise licensing, and VPC deployment.
- Provides publishable compliance information through a trust center, including SOC 2 Type II and HIPAA statements.
- Backs product claims with an evaluations page and customer stories tied to production retrieval workloads.
- Lists clear usage-based pricing for search API components and model endpoints.
Cons
- The site provides limited public detail on integrations beyond the API, SDKs, partner providers, and VPC deployment.
- Pricing and support terms vary by offering, so teams need to review the specific plan or contact sales for enterprise and on-prem arrangements.
- Public materials do not spell out every workflow detail, such as exact ingestion, chunking, or PDF handling behavior, even though the pricing FAQ implies these topics are supported or discussed.
FAQ
What does ZeroEntropy offer?
ZeroEntropy provides specialized AI models for retrieval-heavy workflows, including rerankers, embeddings, and custom models. The pricing page and homepage show both self-serve API usage and enterprise options, including VPC deployment and model licensing.
How can teams integrate or deploy it?
The site shows a ZeroEntropy API with Python and TypeScript SDKs, plus deployment through partner providers and ZeroEntropy VPC. It also notes availability on AWS Marketplace and Azure for VPC deployment.
Is ZeroEntropy positioned for production security and compliance?
Yes. The trust center lists SOC 2 Type II and HIPAA compliance, and the site also references GDPR and CCPA compliance statements. The trust page includes controls for access, data protection, disaster recovery, and network security.
Does ZeroEntropy have enterprise and self-serve options?
The pricing page lists self-serve pricing for the API, an enterprise plan with contact sales flow, and an on-prem offering called "ze on-prem." It also says enterprise customers can get volume discounts, white-glove onboarding, and custom integrations.
Quick Facts
- Category
- Developer Tool
- Primary use
- Retrieval, reranking, and embeddings for AI systems
- Delivery
- API, partner providers, VPC, and enterprise licensing
- Platform notes
- Python and TypeScript SDKs; AWS Marketplace and Azure for VPC deployment
- Trust
- SOC 2 Type II, HIPAA, GDPR, and CCPA statements on the trust center
- Website
- zeroentropy.dev