Cohere launches Embed 5 Pro and Fast for shared-index retrieval

Cohere launched Embed 5 Pro and Fast, two embedding models built around a shared embedding space. Retrieval systems can index documents with the quality-focused Pro model, then query that same index with the lower-latency Fast model without rebuilding it.

This cross-tier setup separates the computational demands of indexing from latency-sensitive searches. Cohere positions Pro, with the model ID embed-v5.0-pro, for offline or quality-critical indexing. Fast, identified as embed-v5.0-fast, is designed for interactive search, agent loops and high-volume queries. The company recommends the Pro-for-indexing, Fast-for-querying pattern for many deployments.

The models bring this index-and-query option to Cohere’s wider enterprise retrieval platform.

Both models accept text, images and combined text-image inputs. According to Cohere’s technical announcement, they support more than 100 languages and a 128,000-token context window.

Cohere says both models are available through its Embed API, Microsoft Foundry and Amazon SageMaker, with Model Vault offered for single-tenant deployment.