Cohere launches Rerank 5 with paragraph-level context focus and calibrated scoring

Cohere launches Rerank 5 with paragraph-level context focus and calibrated scoring

N
News Editor
2026-10-22 17:11:24
Cohere has released its Rerank 5 model family, introducing two versions, Pro and Fast, for retrieval re-ranking before results are passed to a large language model. The company said the new series adds two developer-facing features: Context Focus, which returns relevant passages from within documents, and calibrated relevance scores, which let developers interpret scores as a probability of usefulness. Rerank 5 supports more than 100 languages, a 32K context window, and semi-structured data formats including JSON and tables. Cohere said the Pro version ranked first in 39 of 46 composite datasets in its internal evaluation. On the BRIGHT benchmark using RCP-nDCG@10, it scored 82.2, or 5.8 points above Voyage AI Rerank 3. The Fast version is aimed at low-latency, high-throughput workloads and runs on a single NVIDIA A100. Cohere also said Context Focus is optional and turned off by default. In tests across 30 QA datasets, removing half of the retrieved text preserved 96% of answer quality. In search agent tests, the feature reduced token use per task by 19% to 34% while also improving accuracy, according to MarkTechPost.

Cohere has released the Rerank 5 model family, with Pro and Fast versions designed to reorder retrieval results before they are sent to a large language model.

Two new developer features

The company added Context Focus and calibrated relevance scores. Context Focus returns relevant passages from within a document. Calibrated scoring lets developers read the score as a probability of usefulness.

Rerank 5 supports more than 100 languages, a 32K context window, and semi-structured data including JSON and tables.

How Pro and Fast are positioned

Cohere said the Pro version ranked first in 39 of 46 composite datasets in its internal evaluation. On the BRIGHT benchmark, measured by RCP-nDCG@10, Pro scored 82.2, which was 5.8 points higher than Voyage AI Rerank 3.

The Fast version targets low-latency and high-throughput use cases and runs on a single NVIDIA A100.

Context Focus test results

Context Focus is optional and turned off by default. In tests across 30 QA datasets, Cohere said removing half of the retrieved text still preserved 96% of answer quality.

In search agent tests, using the feature reduced token consumption per task by 19% to 34% while also improving accuracy.

The report was cited by MarkTechPost.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
6300

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.