Product / UX / Agent tools & protocols

Cohere Rerank API

Reranking API that takes a query and a list of documents and returns them reordered by relevance. Purpose-built for RAG pipelines; agents retrieve candidate documents from a vector DB, then call Rerank to get the most relevant ones before feeding them to the LLM. The leading standalone reranking service, used across thousands of production RAG systems. Significantly improves retrieval quality with a single API call.

Cohere Rerank API, verified live June 2026 (Cohere docs). Latest model Rerank 4.0 (fast/pro variants); prior 3.5 and 3.0 (English + multilingual). Proprietary hosted reranking API for RAG/retrieval; also available via Azure AI Foundry, AWS SageMaker/OpenSearch, Oracle GenAI. Sub-segment: retrieval/RAG infra (reranker).

Openness

1 high confidence

1.0

license: proprietary(hosted-API
source: closed
weights: not-released
managed: Cohere+Azure/AWS/Oracle

Proprietary closed reranking models served via API and cloud marketplaces; no open weights or source. Marketed as 'open' in search-relevance contexts but is not OSS.

https://docs.cohere.com/docs/rerank-overview recorded 2026-06-04
commercial Rerank API; models 4.0/3.5/3.0, no source or weight release
https://docs.cohere.com/changelog/rerank-v3.5 recorded 2026-06-04
Rerank v3.5 announced as a hosted proprietary model

Adoption

3 medium confidence

3.0

De facto default managed reranker for RAG, distributed across Azure AI Foundry, AWS SageMaker JumpStart / OpenSearch, and Oracle GenAI in addition to Cohere's API. No hard call-volume/user count published; placed at 3 on broad multi-cloud distribution (reported_traction).

https://aws.amazon.com/blogs/big-data/enhancing-search-relevancy-with-cohere-rerank-3-5-and-amazon-opensearch-service/ recorded 2026-06-04
Rerank 3.5 integrated/distributed via Amazon OpenSearch Service
https://ai.azure.com/catalog/models/Cohere-rerank-v3.5 recorded 2026-06-04
Rerank v3.5 available in Azure AI Foundry model catalog

Capability

4 medium confidence

4.0

Vendor reports SOTA on BEIR and multilingual nDCG@10; widely treated as the reference managed reranker. Concrete nDCG numbers not disclosed on primary pages (claims are vendor-run, not an independent leaderboard), so rated 4 rather than 5. Capability basis is feature_matrix as no standard public reranker leaderboard applies.

https://cohere.com/blog/rerank-3pt5 recorded 2026-06-04
vendor SOTA claims on BEIR + multilingual retrieval, ~25% gain over embedding search
https://docs.cohere.com/changelog/rerank-v3.5 recorded 2026-06-04
multilingual 100+ language reranking, nDCG@10 evaluation

Unchanged since 2026-06-09 (last edited, not re-checked)