Nearly a year after the release of Rerank 3.5, Cohere has launched the latest version of its search model, now with a larger context window to facilitate agents find the information they need to complete their tasks.
Coher said in blog entry that Rerank 4 has a 32K context window, which is a fourfold escalate from 3.5.
“This allows the model to handle longer documents, evaluate multiple sections at once, and capture relationships between sections that would not be possible in shorter windows,” the blog post reads. “Therefore, this increased capacity improves ranking accuracy for realistic document types and increases confidence in the validity of the results.”
Rerank 4 is available in two versions: speedy and professional. As a smaller model, Brisk is best suited for applications 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 more critical this year, especially as AI agents need to access more information and context about the organization they work for. Cohere said reranking tools “significantly improve AI search accuracy in the enterprise by refining initial search results.” Rerank 4 addresses the nuance gap created by some dual-encoder embeddings – models that facilitate search augmented generation (RAG) tasks – by leveraging a multi-encoder architecture “that jointly processes queries and candidates, captures subtle semantic relationships, and reorders results to extract the most relevant elements,” Cohere said.
Performance and benchmarks
Cohere compared these models to other reranking models, such as Qwen Reranker 8B, Jina Rerank v3 from Elasticsearch, and Voyage Rerank 2.5 from MongoDB, for tasks in the finance, healthcare, and manufacturing domains. Rerank 4 performed well, if not better, than its competitors.
Rerank 3.5 stood out for its ability to support multiple languages, and Cohere said Rerank 4 continues that trend. It understands over 100 languages, including cutting-edge search in 10 major business languages.
Reranking agents and models
Rerank 4 aims to enable agents to understand which data best fits their tasks and provide more context.
Cohere noted that the model is a key component of North’s AI agent platform because it “seamlessly integrates with existing AI search solutions, including hybrid, vector and keyword-based systems, with minimal code changes.”
As more enterprises look to exploit agents for research and insights, as evidenced by the rise of Deep Research, models that facilitate filter out irrelevant content, such as reranking tools, become increasingly critical.
“This particularly impacts agent-based AI, where complex, multi-step interactions can quickly trigger 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 everything right, preventing low-quality information from reaching LLM.
Self-education
Cohere said Rerank 4 stands out not only for its extensive rebuilding capabilities, but also for being the first reranking model to learn on its own.
Users can adapt Rerank 4 to the exploit cases they encounter more often, without any additional annotations. Like basic models like GPT-5.2, where people can specify preferences and the model remembers them, Rerank 4 users can inform the model about their preferred content types and document corpora.
For example, when used with Rerank 4 Brisk, the model becomes more competitive with larger models because it is more precise and uses the specific data that users expect.
“Looking further, we also tested how Rerank 4’s machine learning performed in completely new search domains,” Cohere said. “Using healthcare datasets that mimic a clinician’s need to search for patient-specific information – not just medical discipline expertise – we found that enabling self-directed learning resulted in consistent, significant benefits. The result: a clear and significant increase in search quality for Rerank 4 Fast, across the board.”