Cohere’s Rerank 4 quadruples the context window over 3.5 to cut agent errors and boost enterprise search accuracy
Nearly a year after the release of Rerank 3.5, Cohere launched the latest version of its search model, now with a larger context window to help agents find the information they need to complete their tasks.
Cohere said a blog post that Rerank 4 has a 32K context window, which represents a fourfold increase compared to 3.5.
“This allows the model to process longer documents, evaluate multiple passes simultaneously, and capture relationships between sections that shorter windows would miss,” the blog post said. “This expanded capability therefore improves ranking accuracy for realistic document types and increases confidence in the relevance of the retrieved results.”
Rerank 4 is available in two flavors: Fast and Pro. As a smaller model, Fast is best suited for use cases that require both speed and accuracy, such as e-commerce, programming, and customer service. Pro is optimized for tasks that require deeper reasoning, precision, and analysis, such as generating risk models and performing data analysis.
Enterprise search has become increasingly important this year, especially as AI agents need to access more information and context about the organization they work for. Cohere said rerankers “significantly increase the accuracy of business AI searches by refining initial retrieval results.” Rerank 4 addresses the nuance gap created by some bi-encoder embedding models — models that make retrieval augmented generation (RAG) tasks easier — by using a cross-encoder architecture “that jointly processes queries and candidates, captures subtle semantic relationships, and rearranges the results to surface the most relevant items,” according to Cohere.
Performance and benchmarks
Cohere compared the models to other reranking models, such as Qwen Reranker 8B, Elasticsearch’s Jina Rerank v3, and MongoDB’s Voyage Rerank 2.5, for tasks in the financial, healthcare, and manufacturing domains. Rerank 4 performed strongly, if not better than its competitors.
Rerank 3.5 stood out for its ability to support multiple languages, and Cohere said Rerank 4 continues this trend. It understands more than 100 languages, including state-of-the-art querying in 10 major business languages.
Agents and rearrangement models
Rerank 4 aims to help agentic tasks understand which data is most appropriate for their tasks and provide more context.
Cohere noted that the model is a key part of its agentic AI platform North because it “seamlessly integrates into existing AI search solutions, including hybrid, vector and keyword-based systems, with minimal code changes.”
As more companies look to use agents for research and insights, as evidenced by the rise of Deep Research features, models that help filter irrelevant content, such as rerankers, are becoming increasingly important.
“This has a particular impact on agentic AI, where complex, multi-step interactions can quickly drive model calls and saturate context windows,” Cohere said.
The company claims that Rerank 4 helps reduce token usage and the number of retries an agent needs to get things right by preventing low-quality information from reaching the LLM.
Self-learning
Cohere said that Rerank 4 stands out not only for its strong reranking capabilities, but also because it is the first reranking model that learns itself.
Users can customize Rerank 4 for use cases they encounter more often, without additional annotated data. Like basic models like GPT-5.2, where people can indicate preferences and the model remembers them, Rerank 4 users can tell the model their favorite content types and document corpora.
For example, if the model is used with Rerank 4 Fast, it becomes more competitive with larger models because it is more accurate and uses specific data that users want.
“Looking further, we also examined how Rerank 4’s machine learning performs on entirely new search domains,” Cohere said. “Using healthcare-focused datasets that mimic a physician’s need to retrieve patient-specific information (not just expertise from a particular medical discipline), we found that enabling machine learning delivered consistent, substantial gains. The result: a clear and significant improvement in retrieval quality for Rerank 4 Fast, across the board.”
