Cohere Releases Rerank 3.5 with Enhanced Reasoning and Multilingual Capabilities

Dec. 3, 2024 — Cohere has announced the release of Rerank 3.5, its latest AI search foundation model – delivering improved reasoning and multilingual capabilities to help businesses precisely search complex data. Rerank 3.5 is compatible with any existing search system and can be implemented with just a few lines of code. In this recent blog post, Cohere dives into the technical and practical advantages of Rerank 3.5, showcasing how its enhanced reasoning and multilingual capabilities can transform search and retrieval systems across industries.

Rerank 3.5 is our latest AI search foundation model. It enables businesses to significantly improve the relevancy of information surfaced within search and retrieval-augmented generation (RAG) systems.

Rerank 3.5 delivers state-of-the-art features, including:

• Enhanced reasoning skills to enable understanding complex user questions which express various constraints that have traditionally challenged search systems.
• Broad data compatibility to search long documents with rich and relevant metadata (e.g. emails, reports), semi-structured data (e.g. tables, JSON), and code.
• Improved multilingual performance across 100+ languages and industry-leading in global business languages such as Arabic, French, Japanese, and Korean.

“Cohere is a key part of what makes Notion AI work. Their reranker gives us both the speed and quality we need, and it’s consistently improving. It’s been essential for getting our AI Connectors out the door quickly,” said Simon Last, cofounder and CTO at Notion.

Improve Enterprise AI Systems in Minutes

Rerank 3.5 efficiently finds the most relevant business data to answer a user question. It accomplishes this through a method called “cross-encoding” where the model computes a relevance score for a business document in relation to a user question. This method enables highly accurate information understanding, exceeding traditional keyword and embedding search, and can be applied after an initial dense retrieval stage to ensure the answers surfaced to users are maximally precise.

Rerank 3.5 is compatible with any existing search system, can be implemented with just a few lines of code, and tends to have negligible impact to overall system latency. It is most frequently used within retrieval-augmented generation (RAG) applications to provide generative AI models, such as our Command R series, with real-time business context to inform their outputs.

This offers a means of increasing the reliability of AI systems, helping sophisticated businesses move past experimentation and into production deployments to accelerate data-driven decision making.

Understand Complex Data Across 100+ Languages

Search systems often fail to retrieve relevant information when users implicitly or explicitly express constraints on what they would like returned. We identified that this was partially due to traditional systems lacking the ability to reason. Rerank 3.5 shows substantial improvements in this area, understanding complex multifaceted questions that other search systems fail to answer.

This capability is particularly helpful for businesses operating within specialized industries such as finance, government, energy, manufacturing, and healthcare. For example, on a financial services dataset we curated to be generally representative for common use cases, Rerank 3.5 performance was +23.4% better than Hybrid Search and +30.8% better than BM25. We expect organizations in these industries to observe similar improvements when evaluating performance on their data.

Rerank 3.5 also offers industry-leading multilingual capabilities. It can search across data in 100+ languages, with state-of-the-art accuracy on the following 10 global business languages: Arabic, Chinese, French, German, Hindi, Japanese, Korean, Portuguese, Russian, and Spanish.

When compared to our previous Rerank 3 model, Rerank 3.5 delivers a +26.4% improvement on cross-lingual search where the user query is in a different language than the documents being searched. This helps large organizations eliminate barriers to accessing information across geographies and teams.

Get started with Rerank 3.5

Rerank 3.5 is available today on Cohere’s platform, Amazon Bedrock, and Amazon SageMaker. Rerank 3.5 is now for the first time available on Amazon Bedrock through the new Rerank API (learn more). It will soon be available across additional cloud platforms.

Rerank 3.5 can also be deployed into any Virtual Private Cloud (VPC) or on-premise environment. To get started with Rerank 3.5 for your business, and learn more about options for large enterprise deployments, please contact our sales team.

Source: Cohere