Cohere shipped Rerank 5. The update targets how enterprise documents get retrieved correctly, not how well the model generates.
Cohere released Rerank 5 for enterprise document retrieval. The model reorders the candidates a search system returns, ranking them by relevance to the query, and Cohere reports gains in both accuracy and efficiency. This is not a new generative model — it is the component that runs before anything is handed to one. The other items bundled into the same week, a solved mathematical conjecture and a certificate authority compromise, carry no material that connects them to this.
As generative models converge on capability, output quality is decided by what you put in front of them. Rank the wrong document first and the smartest model still answers wrongly. Rerank acts one step earlier, and in enterprise deployments the swing it produces can exceed the choice of generative model. Note: the improvement figures are Cohere's own; no independent comparison has appeared.
Whether third-party benchmarks show Rerank 5 is enough to replace existing stacks. Whether any deployment reports accuracy moving from swapping the retrieval stage alone.