Cohere Rerank 3.5

Cohere Rerank 3.5 is a semantic reranking model built to refine search results within retrieval-augmented generation pipelines. It works as a second-stage filter on top of keyword search or vector similarity, scoring documents against a query to identify which passages most directly address the user's intent. By surfacing the most relevant context for a given question, it helps reduce noise and lower token costs before generation, making it well suited for enterprise search, question answering, and knowledge-base applications that rely on document retrieval.

The model has been adopted by multiple serving platforms, signaling broad ecosystem integration for production reranking workloads. Cohere Rerank 3.5 became available on Amazon Bedrock through the Rerank API, as announced on the AWS Machine Learning Blog, allowing developers to incorporate high-fidelity semantic ranking into Bedrock-hosted applications. It was also listed among the reranking models made generally available on Heroku Managed Inference and Agents, where the Heroku Rerank API is designed to be compatible with the Cohere request format, easing adoption for teams already running retrieval pipelines in that environment.