Embed 4
Embed 4 remains listed for text, image and mixed-document embeddings. It offers selectable vector dimensions and a 128K input context, with 1536 dimensions as the default.
Model behavior
This is the preceding embedding generation. Cohere's catalog continues to list it alongside Embed 5. The API produces vector representations for search rather than written answers.Cohere model catalog
How to use it well
Practical guidance. An existing Embed 4 index is a reason to review migration carefully. Compare retrieval on representative queries before rebuilding the whole archive. Keep the encoder version attached to the index configuration and avoid mixing vectors from different generations without a documented compatibility claim. For a new project, include Embed 5 in the comparison, especially when images and document layout carry information that plain text loses.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | embed-v4.0 |
| Context window | 128,000 tokens Published maximum input context. Cohere Embed specifications |
| Vector dimensions | 256, 512, 1024 or 1536Cohere Embed specifications |
Related models
- Embed 5 Fast
Cohere · Embed
Low-latency embedding model that shares its vector space with Embed 5 Pro.
Launch date not confirmedAPI - Embed 5 Pro
Cohere · Embed
Quality-focused embedding model for text, images and mixed documents, sharing a vector space with Embed 5 Fast.
Launch date not confirmedAPI - Embed Multilingual 3
Cohere · Embed
Multilingual text embedding model for retrieval across languages.
Launch date not confirmedAPI