DEVOURED
Announcing Cohere's Embed 5 Models
Cohere has released Embed 5, offering significant improvements in retrieval performance for complex documents, multilingual text, and mixed-mode PDF analysis.
What: Embed 5 comes in 'pro' and 'fast' versions, both sharing a common embedding space. It features 128k context windows, support for 100+ languages, and Matryoshka dimensionality for flexible vector storage.
Why it matters: Embedding models are increasingly differentiated by their ability to handle multi-modal, long-context data rather than just pure text, which is critical for RAG (Retrieval-Augmented Generation) at scale.
Takeaway: Switch your indexing model to 'embed-v5.0-pro' for high-accuracy offline search and 'embed-v5.0-fast' for production query traffic.
Deep dive
- Shared embedding space allows mixed-model architectures.
- Supports Matryoshka dimensions (256 up to 2048) for storage optimization.
- Enhanced performance for visually rich documents, financial filings, and code repositories.
- Available via Cohere API, Microsoft Foundry, and Amazon SageMaker.
Decoder
- Matryoshka embeddings: A technique for training embeddings where a single vector can be truncated to smaller dimensions while retaining meaningful semantic information, allowing for tiered search performance.
- RAG (Retrieval-Augmented Generation): The process of retrieving external data and providing it as context to a model to improve accuracy and grounding.
Original article
Announcing Cohere's Embed 5 Models
We’re pleased to announce the release of Embed 5, Cohere’s most powerful embeddings family yet.
Embed 5 delivers frontier retrieval quality on complex enterprise data, with major gains over Embed 4 on visually rich documents, financial filings, parsed PDFs, code, and multilingual retrieval.
Key features
- Two model variants available:
embed-v5.0-pro: Optimized for the highest retrieval quality, particularly for offline indexing and quality-critical retrievalembed-v5.0-fast: Optimized for low latency and high throughput, particularly for interactive search, agent loops, and high-volume query traffic
- Shared embedding space: Pro and Fast share an embedding space, so a corpus indexed with one model can be queried with the other. We recommend indexing with Pro and querying with Fast.
- Multimodal inputs: Embed text, images, and mixed text-and-image inputs (e.g. PDF pages) in a single vector
- Multilingual support: Supports over 100 languages
- Extended context length: 128k token context window
- Flexible storage: Matryoshka embeddings in the following dimensions:
[256, 512, 768, 1024, 1536, 2048], withfloat,int8, andbinaryoutput types
Availability
Embed 5 is available through the Embed API, as well as Microsoft Foundry (Pro, Fast) and Amazon SageMaker (Pro, Fast). For single-tenant deployment, Embed 5 is also available in Cohere’s Model Vault.
For more details, see the model documentation.
Explore all 57 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi