Cohere released Rerank 5 in Pro and Fast versions for general availability. Its new Context Focus feature identifies and returns relevant text passages while ranking documents, allowing less context to be sent to a downstream LLM. A higher threshold can reduce input further but may also remove information needed for the answer. In its own evaluations, the company reported little loss of answer quality after removing half the context, while Fast prioritizes throughput and latency. These figures reflect particular evaluations and generation models, so accuracy and costs should be measured together on the search data actually used.