Cohere Releases Rerank 5 With Passage-Level Context Focus and Calibrated Scores

Cohere has launched Rerank 5, its latest document reranking model designed to improve retrieval quality in search and RAG pipelines. The model is available in two tiers—Pro and Fast—each supporting a 32K token context window, which allows it to process longer documents and multi-passage queries in a single pass.

A key feature of Rerank 5 is Context Focus, a passage-level extraction mechanism that identifies the most relevant segments within a document rather than scoring the entire text uniformly. This approach can improve precision when documents contain mixed relevant and irrelevant content, a common challenge in enterprise search and retrieval-augmented generation systems.

The model also introduces calibrated relevance scores, meaning the output scores are more directly interpretable as probability-like values. This calibration can simplify threshold tuning for downstream applications such as filtering, ranking cutoffs, and confidence-based routing.

On the BRIGHT benchmark, the Pro tier achieves a score of 82.2, while pricing is set at $0.05 per million tokens. For teams building search or RAG systems, the combination of a larger context window, passage-level focus, and calibrated outputs positions Rerank 5 as a practical upgrade over earlier reranking models, particularly for applications where retrieval quality directly affects answer accuracy.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.