Cohere Rerank v4.0
used in
0
stacks
Cohere's current-generation reranking models (Pro and Fast variants) for boosting retrieval relevance in RAG and agent search.
Rerank v4.0 is Cohere's latest reranking generation, sorting text inputs by semantic relevance to a query. It ships in two variants, a Pro model optimized for quality on complex scenarios and a Fast model for low-latency, high-throughput use cases; both are multilingual and handle semi-structured data such as JSON. Documents exceeding the context limit are automatically chunked, and the models are available API-only via the Rerank endpoint, with v3.5 and v3.0 still offered.
Strengths
- Purpose-built second-stage reranking improves precision of any retriever
- Multilingual and handles semi-structured JSON documents
- Pro/Fast split lets builders trade quality against latency and throughput
Tradeoffs
- API-only (no open weights)
- Adds an extra network hop and per-query cost to the retrieval pipeline