Cohere Unveils Rerank 3: A New Foundation Model for Enterprise Data Search and RAG

Cohere Unveils Rerank 3: A New Foundation Model for Enterprise Data Search and RAG

Cohere, a startup providing enterprise AI models and solutions, has launched its newest foundation model, Rerank 3, designed to enhance enterprise search and retrieval systems. Rerank 3 drastically improves the efficiency and accuracy of retrieval augmented generation (RAG) systems and is tailored to handle the complex and diverse datasets that enterprises typically manage.

Enterprises must deal with complex, multi-aspect, and semi-structured data across various formats, including emails, invoices, JSON documents, code, and tables. However, with a single line of code, enterprises can seamlessly integrate Rerank 3 into their existing search systems or legacy applications, significantly boosting search performance while minimizing latency and cost. Here's a breakdown of what the Rerank 3 model offers:

  • Multi-Aspect Data Handling: Rerank 3 excels at searching through multi-aspect data sources like emails, invoices, and other complex data structures. It can consider multiple metadata fields, including recency, to provide relevant results.
  • Code and Documentation Retrieval: The model offers improved code retrieval capabilities, enabling enterprises to search their proprietary code repositories or vast documentation efficiently.
  • Multilingual Support: With coverage of over 100 languages, Rerank 3 simplifies retrieval for global organizations dealing with multilingual data sources, enhancing accessibility for non-English speaking users.
  • Longer Context, Improved Accuracy: Featuring a 4k context length, Rerank 3 can process larger documents, reducing the need for chunking. This enhances search quality, especially for long-form content, ensuring more context is considered for relevance scoring.
  • Enhanced Table and JSON Data Search: Rerank 3 introduces the ability to search semi-structured data in JSON format and improves tabular data search, allowing enterprises to tap into critical information in databases, CSVs, and Excel sheets.
(left) Retrieval augmented generation workflow without Rerank. Documents are retrieved from an existing search system and are passed directly to the LLM for grounded generation. (right)

Rerank 3 also offers significant advantages over traditional approaches to RAG systems:

  • Reduced Latency, Improved Accuracy: Rerank 3 boosts RAG performance by isolating the most relevant documents for a user's query. This not only enhances response accuracy but also lowers latency, making it ideal for time-sensitive applications.
  • Cost Savings: When combined with Cohere's Command R family of models, Rerank 3 reduces the total cost of ownership (TCO) for RAG systems by up to 98% compared to other generative LLMs, without compromising on quality.
  • Superior Ranking Accuracy: Rerank 3 outperforms industry-leading LLMs in ranking accuracy while being significantly more cost-effective, making it the go-to choice for efficient and effective RAG implementations.

Cohere has also announced native support for Rerank 3 in Elastic's Inference API, enabling seamless integration with Elasticsearch, a widely adopted search technology. This partnership will allow developers to easily build robust enterprise search systems, leveraging the strengths of both platforms.

With Rerank 3, Cohere has delivered a powerful tool for enterprises to harness the power of their data. By providing efficient, accurate, and multilingual search capabilities, businesses can access and leverage information from various sources, enabling data-driven decision-making and unlocking new possibilities for growth and innovation.

Chris McKay is the founder and chief editor of Maginative. His thought leadership in AI literacy and strategic AI adoption has been recognized by top academic institutions, media, and global brands.

Let’s stay in touch. Get the latest AI news from Maginative in your inbox.

Subscribe