Rerank 5 Pro
AI model · Reranker / retrieval · Text → Tables / structured data by Cohere
Sharper search, fewer tokens, and nuanced relevance for enterprise retrieval.
Cohere's premier reranking model for enterprise search and RAG, featuring Context Focus for passage-level relevance and support for 100+ languages.
Model specs
- Provider
- Cohere
- Model family
- Rerank 5
- Released
- 2026-10-21
- Context window
- 32,000 tokens
- Input
- Text, Image
- Output
- Text
- Search unit
- 1 query with up to 100 documents ranked
- Capabilities
- Structured output
- Available on
- API, Amazon SageMaker, Microsoft Foundry, Model Vault, OpenRouter
- Launched
- 2026-10-21
- Languages
- English, French, Spanish, German, Chinese, Japanese
Category: Natural Language Processing
- Modalities
- Text → Text
- Context window
- 32,768 tokens
- Sizes and variants
- Rerank 5 Pro, Rerank 5 Fast
- Released
- 2026-10-21
- Languages
- 100+ languages
- Supported Languages
- 100+
- 32,768 token context window
- Context Focus for passage-level relevance retrieval
- Supports 100+ languages
- Available on API, Model Vault, SageMaker, and Azure
Overview
It serves as a final stage in the search pipeline, re-ordering results from initial retrieval systems based on nuanced semantic relevance to the user's query. The model is specifically optimized for reasoning-intensive retrieval in specialized domains such as finance, healthcare, and scientific research. A key feature is Context Focus, which identifies and returns the most relevant character spans within long documents to reduce token costs for downstream LLMs. Rerank 5 Pro supports over 100 languages and a 32,768 token context window. It is available through Cohere's API, dedicated Model Vault infrastructure, and major cloud platforms including Amazon SageMaker and Microsoft Azure.
Ranking and Relevance
- Relevance Optimization: The model acts as a final stage in search pipelines, re-ordering documents based on nuanced relevance to the user's query.
- Reasoning-Intensive Retrieval: Rerank 5 Pro is optimized for complex, reasoning-heavy queries and long or domain-specific documents.
- Semi-structured Data Support: The model supports ranking for JSON, tables, and other records formatted as YAML strings.
- Calibrated Relevance Scores: Relevance scores are consistent across queries and domains, allowing users to set a single threshold for filtering results.
Context Focus Features
- Context Focus Passage Retrieval: This optional capability identifies and returns the most relevant character spans within each document.
- Token Cost Reduction: By identifying specific relevant passages, the model helps reduce the number of tokens sent to downstream LLMs.
- User-Facing Citations: Users can highlight the specific passages returned by the model to show why a result matched a query.
Deployment and Access
- Cohere API Access: The model is available through the Cohere Rerank API for seamless integration.
- Model Vault Deployment: Enterprises can deploy the model on dedicated, fully managed infrastructure through Cohere Model Vault.
- Cloud Marketplace Availability: Rerank 5 Pro is available as a managed service on Amazon SageMaker and Microsoft Azure Foundry.
- OpenRouter Integration: The model is accessible via OpenRouter for developers using their existing OpenRouter API keys.
Use cases
- Financial analysts use Rerank 5 Pro to surface precise information from complex filings and research reports for investment decision-making.
- Healthcare researchers rank scientific papers and clinical data to find the most relevant passages for medical inquiries.
- Legal teams refine search results across large document corpora to identify specific clauses and relevant case law.
- Developers integrate Rerank 5 Pro into RAG pipelines to reduce LLM token costs by passing only the most relevant document spans.
Specifications
- Modalities
- Text → Text
- Context window
- 32,768 tokens
- Sizes and variants
- Rerank 5 Pro, Rerank 5 Fast
- Access
- API, Model Vault, Amazon SageMaker, Microsoft Azure, OpenRouter
- Released
- 2026-10-21
- Languages
- 100+ languages
- Supported Languages
- 100+
- Input Modality
- Text
- Output Modality
- Relevance scores and character spans
- Data Formats
- Raw text, JSON, YAML, Tables
- Deployment Platforms
- API, Model Vault, SageMaker, Azure, OpenRouter
How it works
- Integration: Integrate the Rerank 5 Pro model into your existing search or RAG pipeline via the Cohere API or a cloud provider.
- Query Processing: Send a query and a list of documents to the Rerank endpoint; the model evaluates the semantic relevance of each document to the query.
- Passage Extraction: Optionally enable Context Focus to receive specific relevant character spans within each document to reduce downstream token usage.
- Result Refinement: The model returns a list of documents sorted by relevance score, which can then be passed to a generative LLM or displayed to the user.
Pricing
- Search Unit (1 query, up to 100 docs)
- Contact sales for production pricing
FAQ
Do I need to re-index my data to upgrade to Rerank 5?
No, Rerank reads raw text at query time, so upgrading from an earlier model only requires updating the model parameter to rerank-v5.0-pro without re-indexing your corpus.
What languages does Rerank 5 Pro support?
Rerank 5 Pro supports over 100 languages for global enterprise retrieval.
Added by AntSpace from public sources (cohere.com).