Cohere lets fast queries search the same Pro index
Cohere released Embed 5 Pro and Fast, two models that turn documents and queries into compatible search vectors. A collection indexed with Pro can be queried with Fast without rebuilding it. Both accept text and images. The release separates the work of preparing a document collection from the repeated searches that use it.
Artificial Intelligence··Midday
Pro and Fast share one search index
Cohere released Embed 5, a family of models that converts text and images into vectors used to find related material. Its Pro and Fast variants share a compatible vector space. Documents prepared with Pro can therefore be searched with Fast without building a second index.[1], [2]
Cohere positions Pro for indexing that prioritizes retrieval quality and Fast for latency-sensitive searches and high-volume queries. The shared representation keeps the prepared collection usable when a team changes the model handling its queries.[1]
Text and images enter the same workflow
Both variants accept text, images and documents that combine the two. Cohere lists more than 100 languages and a context window of 128,000 tokens, the units a model uses to process input. A visually rich document can enter the search workflow alongside ordinary text.[1]
Shorter vectors change storage requirements
Teams can select vector dimensions from 256 to 2,048 and use floating-point, int8 or binary outputs. Shorter representations are intended to reduce storage and search costs. The models are available through Cohere’s API and managed platforms, with private deployment supported. Retrieval and throughput comparisons remain measurements from Cohere’s own evaluation setup.[1]
