Cohere Launches Embed 5 Enterprise Retrieval Model Family

01 What happened

Cohere released the Embed 5 family of models for enterprise retrieval tasks. The models are available via the Cohere API, Model Vault, Microsoft Foundry, and Amazon SageMaker.

02 Key details

• Embed 5 supports a 128K-token context window and processes over 100 languages.
• The Pro and Fast tiers share a single embedding space, allowing users to index with Pro and query with Fast without rebuilding the index, provided output dimensions and compression settings like Matryoshka and int8 match.
• Pricing is $0.12 per million tokens for Pro and $0.08 per million tokens for Fast; image processing costs $0.40 per million tokens.
• Cohere reports that Embed 5 Pro achieved an average score of 85.8 on the ViDoRe V3 benchmark.

03 Why it matters

Shared embedding spaces allow organizations to balance retrieval performance and cost by using Pro for indexing and Fast for queries.

04 Who it matters to

Corporate AI platform developers, data engineers, teams working with internal data and enterprise AI customers.