Cohere Releases Embed 5 Models with Multimodal Support and 128k Context

Cohere has released Embed 5, its most powerful embeddings family yet, designed to deliver frontier retrieval quality on complex enterprise data. The models show major gains over Embed 4 on visually rich documents, financial filings, parsed PDFs, code, and multilingual retrieval.

Model Variants and Architecture

The release includes two variants optimized for different use cases:

• embed-v5.0-pro: Optimized for the highest retrieval quality, particularly for offline indexing and quality-critical retrieval.
• embed-v5.0-fast: Optimized for low latency and high throughput, particularly for interactive search, agent loops, and high-volume query traffic.

Pro and Fast share a shared embedding space, allowing a corpus indexed with one model to be queried with the other. Cohere recommends indexing with Pro and querying with Fast.

Key Capabilities

Embed 5 introduces several technical enhancements:

• Multimodal inputs: Supports text, images, and mixed text-and-image inputs (e.g., PDF pages) in a single vector.
• Multilingual support: Covers over 100 languages.
• Extended context length: Features a 128k token context window.
• Flexible storage: Offers Matryoshka embeddings in dimensions [256, 512, 768, 1024, 1536, 2048] with float, int8, and binary output types.

Availability

Embed 5 is available through the Embed API, Microsoft Foundry, and Amazon SageMaker. For single-tenant deployment, it is also accessible via Cohere’s Model Vault.