Cohere Embed 5 and Rerank 5 Land on OpenRouter

Cohere has put its newest search models, Embed 5 and Rerank 5, on OpenRouter, a service that sells access to many companies' models through one API and one bill. Cohere's announcement names the two models and their Pro and Fast tiers but does not say what they do. Both belong to retrieval: the part of an AI application that finds the documents a question should be answered from.

The two models do different jobs. An embedding model turns each document and each query into a list of numbers, so a search system can return the documents whose numbers sit closest to the query's. Cohere's Embed 5 documentation says it accepts text, images and mixed pages such as PDF pages. A reranker runs after that first search. According to the Rerank 5 documentation, it takes the results an existing keyword, vector or hybrid search returned, reorders them by relevance to the query and gives each a score between 0 and 1. Cohere suggests using those scores to set relevance thresholds and to choose which passages to send to a language model.

The tiers trade quality for speed. Cohere describes Pro as the highest-quality version of each model and Fast as the choice for low-latency, high-volume work such as interactive search and agent loops. For embeddings, the two tiers share one embedding space, so a collection indexed with one can be searched with the other. Cohere recommends indexing with Pro and querying with Fast, which puts the slower model on the one-time indexing work and the faster one on every incoming query. The announcement's "state-of-the-art" claim comes without benchmarks, and this explanation does not verify Cohere's quality claims or OpenRouter's prices.