↓

Model Cohere ·

Introducing Rerank 5: precise, efficient enterprise document retrieval

Cohere has released Rerank 5 in Pro and Fast tiers, adding a "Context Focus" feature that pinpoints the most relevant passages and relevance scores calibrated as probabilities. The company reports a score of 82.2 on the BRIGHT benchmark for Pro.

What the article says

Cohere introduces Rerank 5, its latest reranking model for enterprise search, in a post dated October 21, 2026. It comes in two tiers, Pro (ranking quality) and Fast (latency and cost), and joins Embed 5 and Parse 5. Cohere positions it for RAG and agentic workflows, where retrieval quality determines how many tokens an LLM reads.

Three points stand out: Context Focus, an optional feature that returns scored text spans within each document; calibrated relevance scores, presented as the probability that a document is useful so one threshold works across queries; and performance that Cohere says leads its benchmarks. It works on top of existing search without re-indexing, using a relevant_spans parameter. It is available through the Cohere API, Model Vault, Microsoft Foundry and Amazon SageMaker, priced at $0.05 (Pro) and $0.02 (Fast) per million tokens; the performance figures are Cohere's own.

Read the official article AnnuIA profile

Summary written by AnnuIA in its own words from the official page of Cohere. The content and rights belong to the publisher: please read the original article.

Latest AI news

All news