Cohere releases Rerank 5 with passage-level context focus and calibrated scores

Cohere reports that the Pro tier ranks first on 39 of 46 composite datasets and scores 82.2 on the BRIGHT benchmark, 5.8 points above Voyage AI Rerank 3. Median latency is 11.7 s per 96‑document request versus 4.1 s for Fast. In internal tests, Context Focus cut tokens by 19 % to 34 % while keeping 57 %‑58 % answer accuracy, and reduced tokens by 41 % on deep‑search benchmarks with accuracy rising from 39 % to 52 %. Calibrated scores achieve an expected calibration error of about 0.06. Pricing is $0.05 per 1 M tokens for Pro and $0.02 for Fast, with Model Vault instances at $4.00 per hour. Integrations include LangChain, Haystack, Elasticsearch and Weaviate. All benchmark figures are self‑reported by Cohere and have not been independently verified.