Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now
Cohere, a leading provider of natural language processing solutions, has announced the release of Rerank 3, a state-of-the-art foundation model designed to revolutionize enterprise search and retrieval. This new offering promises to significantly enhance the ability of businesses to extract valuable insights from complex data sources, such as JSON, emails, and tables, across multiple languages.
Nils Reimers, Director of Machine Learning at Cohere, explained the unique capabilities of Rerank 3 in an interview with VentureBeat. “Searching through complex data sources like JSON, emails, and tables which are widely used across enterprises has historically been very difficult,” Reimers said. “Rerank 3 is different — it can disentangle all aspects that are mentioned in the input, and represent these independently. This makes it much better at complex enterprise data like JSON, emails and tables.”
One of the key features of Rerank 3 is its extended context length of 4,000 tokens, which dramatically improves search quality for longer documents and reduces the need to split up data. “Previous approaches for search often only worked up to 300 words, making it difficult to find long relationships in text to connect information,” Reimers told VentureBeat. “Rerank 3 has been carefully trained and tested to be able to connect information up to 4k tokens. This offers a significant improvement especially for more complex questions, where the answer cannot be inferred from a single paragraph.”

Seamless integration and partnership with Elastic
The integration of Rerank 3 with Retrieval Augmented Generation (RAG) systems is another major advantage, enabling improved response accuracy and cost efficiency for various enterprise use cases. “Rerank 3 ensures that the most relevant documents appear first, requiring users to pass less context information to the LLM, and offering a much quicker and more cost efficient response,” Reimers explained.
AI Scaling Hits Its Limits
Power caps, rising token costs, and inference delays are reshaping enterprise AI. Join our exclusive salon to discover how top teams are:
- Turning energy into a strategic advantage
- Architecting efficient inference for real throughput gains
- Unlocking competitive ROI with sustainable AI systems
Secure your spot to stay ahead: https://bit.ly/4mwGngO
Cohere has also partnered with Elastic to make Rerank 3 natively supported in Elastic’s Inference API, allowing developers using Elasticsearch to seamlessly benefit from enhanced reranking capabilities. “Developers with data stored in existing Elasticsearch indexes will benefit from our enhanced reranking capabilities. We’re excited to deepen our partnership with Elasticsearch and look forward to working together to deliver more powerful enterprise solutions,” Reimers said.

Unraveling the complexities of enterprise data
As foundation models like Rerank 3 become increasingly essential for enterprises, Cohere remains committed to responsible development and use of these technologies. “Cohere maintains a strong commitment to data privacy and security for our enterprise customers. We’ve always built products with data privacy at the core,” Reimers told VentureBeat.
With the release of Rerank 3, Cohere continues to solidify its position as a frontrunner in the rapidly evolving field of natural language processing and enterprise AI solutions. The company’s innovative approach to tackling complex data challenges and its dedication to responsible AI development make it a compelling choice for businesses seeking to harness the power of advanced search and retrieval technologies.
As Reimers aptly put it, “We are excited about the possibilities Rerank 3 enables on semi-structured and tabular data…This new capability of Rerank 3 unlocks massive opportunities for enterprises.” And with Cohere at the helm, those opportunities are closer than ever before.