Cohere launches rerank 5 with enhanced contextual relevance and efficiency

Cohere AI has launched Rerank 5, a new retrieval model designed to improve search result ranking before they are processed by large language models. The model is available in two tiers: Pro and Fast. Pro focuses on high-quality ranking for complex queries and long documents, while Fast prioritizes low latency and high throughput for agents performing multiple searches. Both models support over 100 languages and a 32K context window, and can handle semi-structured data like JSON and YAML. Rerank 5 introduces Context Focus, a feature that highlights relevant passages within documents, and calibrated relevance scores. Testing showed that reducing retrieved text by up to 75% only slightly impacted answer quality, with 87% remaining intact. In benchmark tests, Rerank 5 improved accuracy by up to 12 percentage points compared to BM25, while reducing token usage by 35%. Cohere claims Rerank 5 outperforms competitors on key benchmarks, with Pro leading in ViDoRe V3, BEIR, BRIGHT, and nanoMTEB. The model is now available via Cohere’s API, Model Vault, Microsoft Foundry, and Amazon SageMaker, with pricing starting at $0.02 per 1M tokens for the Fast tier.