Cohere Rerank API

weights

closed(no weights download, Hub card or self-hosted build for any model in the line)

data

closed(no training corpus published)

code

closed(hosted API only

license

proprietary(hosted API, billed in search units, also resold through Azure, AWS and Oracle)

Cohere Rerank is offered only as a hosted API and through cloud marketplaces, with no downloadable weights, no published training data and no way to run it outside Cohere's service. That closes off every higher reading regardless of the license terms attached to any tier.

• https://docs.cohere.com/changelog/rerank-v3.5 recorded 2026-09-11

  Release notes for Rerank 3.5 as a hosted model reached through the API, with a context length of 4096. No weights release accompanies it.

• https://docs.cohere.com/docs/rerank-overview recorded 2026-09-11

  Model lineup - "Rerank 4.0 (both 'fast' and 'pro'): A single multilingual model (rerank-v4.0-pro and rerank-v4.0-fast). Rerank 3.5: A single multilingual model (rerank-v3.5). Rerank 3.0: Separate English-only and multilingual models." Calls are billed in search_units. No weights download, hub model card or self-hosted build for any of them.

Cohere publishes no call-volume or user figures for Rerank, so adoption is read from distribution: the API is offered directly and resold as a managed service on Azure AI Foundry, AWS SageMaker JumpStart and Oracle GenAI. The `cohere` PyPI package bundles chat, embedding, classification and reranking together, so its downloads are not a proxy for Rerank alone.

• https://ai.azure.com/catalog/models/Cohere-rerank-v3.5 recorded 2026-09-28

  Azure AI Foundry catalog, Cohere provider entry: "high‑precision Rerank APIs, Cohere focuses on enterprise and regulated‑industry use cases" and "with Azure AI Search, store embeddings in Azure Vector DBs". It establishes availability and carries no usage figure.

• https://aws.amazon.com/blogs/big-data/enhancing-search-relevancy-with-cohere-rerank-3-5-and-amazon-opensearch-service/ recorded 2026-09-28

  AWS Big Data Blog post "Enhancing Search Relevancy with Cohere Rerank 3.5 and Amazon OpenSearch Service", on wiring the model in through "OpenSearch Service ML connectors which facilitate access to models hosted on third-party ML platforms". It carries no usage figure.

Cohere reports state-of-the-art BEIR and multilingual results for Rerank 3.5 and says Rerank 4 beats Voyage and Jina rerankers on its own benchmarks, and Agentset's independent leaderboard places Rerank 4 Pro second, just behind Zerank 2 and ahead of Voyage rerank-2.5. That leaderboard ranks models by LLM-judged head-to-head wins on only three datasets, so it places Rerank 4 at the competitive frontier alongside those rerankers rather than clearly ahead of them.

• https://agentset.ai/rerankers recorded 2026-09-28

  Agentset reranker leaderboard, last updated February 15, 2026: Cohere Rerank 4 Pro ranks second on ELO, behind Zerank 2 and ahead of Voyage AI Rerank 2.5, over "FiQA (finance), SciFact (science), and PG (long-form content)", with PG "evaluated only with ELO-based LLM judgments".

• https://blog.voyageai.com/2025/08/11/rerank-2-5/ recorded 2026-09-28

  Voyage rerank-2.5 launch post: "On our standard suite of 93 retrieval datasets spanning multiple domains, rerank-2.5 and rerank-2.5-lite improve retrieval accuracy by 7.94% and 7.16% over Cohere Rerank v3.5". It predates Rerank 4.

• https://cohere.com/blog/rerank-3pt5 recorded 2026-09-28

  Rerank 3.5 launch post: "Multilingual performance is measured by nDCG@10" over "external datasets covering 18 different languages in a variety of monolingual and cross-lingual settings", and against Rerank 3 it reports a "+26.4% improvement on cross-lingual search". Searched the body for BEIR: none.

• https://cohere.com/blog/rerank-4 recorded 2026-09-28

  Rerank 4 launch post: a 32K context window, "the largest of our Rerank series to date and a four-fold increase over the previous generation", with Cohere claiming it is "outperforming the likes of MongoDB’s Voyage models and ElasticSearch’s Jina rerankers in overall search relevance". The comparison is Cohere's own.

• https://docs.cohere.com/changelog/rerank-v3.5 recorded 2026-09-28

  Changelog entry: "Rerank 3.5 has a context length of 4096, SOTA performance on Multilingual Retrieval tasks and Reasoning Capabilities" and "Rerank 3.5 has SOTA performance on BEIR and domains such as Finance, E-commerce, Hospitality". No nDCG figure is given here.

• https://docs.cohere.com/docs/rerank-overview recorded 2026-09-28

  Rerank docs overview: "Cohere’s Rerank models have been trained for performance across 100+ languages." The lineup lists Rerank 4.0 in "(both ‘fast’ and ‘pro’)" variants alongside Rerank 3.5.