Sign in
Large Language ModelText GenerationCohere Rerank Family

Cohere Rerank

Cohere Rerank is a production search and retrieval reranking model that scores and reorders candidate documents based on their relevance to a query. It is optimized for information retrieval, semantic search, and question answering pipelines rather than general-purpose text generation.

by CohereProprietary
API Access
Available

Key Capabilities

  • +Learning-to-rank style reranking of search results
  • +Optimized for semantic search and retrieval-augmented generation (RAG) pipelines
  • +Supports multiple domains such as web, enterprise, and e-commerce search
  • +High-throughput, low-latency API suitable for production search systems

Limitations

  • -Not a general-purpose generative LLM; focused on scoring and ranking
  • -Benchmarks are typically retrieval-focused (e.g., MTEB) rather than standard LLM benchmarks like MMLU or HumanEval
  • -Requires an existing set of candidate documents to rerank; does not perform retrieval itself

Benchmark Performance

embedding

embeddingsource

Massive Text Embedding Benchmark

64.5%
embedding

MTEB Retrieval Average

55.0%

Alternatives & Comparisons

OpenAI text-embedding-3-large + custom rerankerretrieval/reranking

Requires building your own reranking layer on top of embeddings rather than using a dedicated rerank model API.

Strengths
  • + Highly flexible
  • + Strong embedding quality
Weaknesses
  • - More engineering effort
  • - No dedicated rerank scoring out of the box
Voyage AI rerank modelsretrieval/reranking

Competing hosted rerank APIs focused on search and RAG quality.

Strengths
  • + Strong retrieval benchmarks
  • + Simple API
Weaknesses
  • - Smaller ecosystem than Cohere
  • - Pricing and SLAs may differ

Sources

docs.cohere.comcohere.com