Use Cohere Rerank 4 in OCI Generative AI

You can now rerank documents with Cohere Rerank 4 in OCI Generative AI. Rerank 4 improves enterprise retrieval workflows with a larger context window, stronger reranking quality, self-learning support, and two variants optimized for different workload needs.

Depending on the model and region, Cohere Rerank 4 is available on-demand and through dedicated AI clusters.

Key Features

• Larger context window: Supports a 32,000-token context window for long documents and larger candidate inputs.
• Improved reranking quality: Improves result ordering for enterprise retrieval workloads, including business, finance, and technical content.
• Self-learning support: Helps adapt reranking behavior to your data, terminology, and relevance preferences without annotated training data.
• Two model variants:
  • Cohere Rerank 4 Pro for higher-precision and more complex reranking tasks.
  • Cohere Rerank 4 Fast for lower-latency and higher-throughput workloads.
• Multilingual support: Supports reranking for English and non-English content across more than 100 languages.
• Semi-structured data support: Supports JSON, tables, and code-like content.

Model Names

• cohere.rerank-v4.0-pro
• cohere.rerank-v4.0-fast

Model Access

You can access these models only through API. Call the RerankText API from a supported region. In RerankTextDetails, for servingMode, set the servingType attribute to ON_DEMAND for on-demand access or DEDICATED for a model hosted on a dedicated AI cluster endpoint.

For available regions, and whether each variant is offered on-demand or through dedicated AI clusters, see Generative AI Models by Region. For model details, see Cohere Rerank 4. For information about the service, see the Generative AI documentation.