Email content

To view this email as a web page, click here

Logo

Embed5_Hero_Email.png

We are excited to announce that

Embed 5 is now generally available. It is Cohere’s most powerful embedding family yet, delivering frontier retrieval accuracy on complex, multimodal and multilingual enterprise data.

Embed 5 excels with real-world document corpora. It comes in two tiers that share a single embedding space:

Embed 5 Pro is optimized for the highest retrieval quality and

Embed 5 Fast brings highly competitive performance to latency- and cost-sensitive workloads at a third of the price. Teams can index with Pro and query with either model on the same index.

Key highlights:

Frontier retrieval on real enterprise data: Embed 5 Pro leads the market on visually rich documents, financial retrieval, parsed PDFs, and key business languages, with a clear step up from Embed 4.

A Fast tier that leads its class: Embed 5 Fast leads other compact models, and goes toe-to-toe with larger general-purpose embeddings on a range of enterprise domains (e.g. financial services).

One embedding space, two operating points: Pro and Fast can be used interchangeably, including with truncated and lower-precision vectors.

Built for how enterprises store and serve retrieval: Multimodal inputs, 100+ languages, a 128K context window, flexible embedding dimensions, and quantized outputs with near-full-precision quality.

We built Embed 5 for enterprise search, RAG, and agentic workflows, where strong retrieval means better experiences for users and less spend on irrelevant context.

This is also the first model family optimized for Cohere’s new retrieval accuracy methodology - designed to better reflect real-world enterprise search and backed by industry leaders.

Learn more here.

Embed 5 is generally available today on the

Cohere API,

Model Vault, Microsoft Foundry

(Pro &

Fast), and Amazon SageMaker

(Pro &

Fast), or for self-hosting deployments.

Create an API key now to start indexing your data. To learn more,

read the announcement.