Cohere embed-v4
FreeMultilingual, int8 support
FreeFree tier
Inputs: text, image
About Cohere embed-v4
Cohere's Embed v4.0 is a state-of-the-art embedding model that converts text, images, and mixed content (like PDFs) into dense vector representations. It supports over 100 languages, offers flexible output dimensions (256, 512, 1024, or 1536), and handles up to 128k tokens of context. The model enables semantic search, classification, clustering, and other similarity-based tasks using cosine similarity, dot product, or Euclidean distance. It is designed for high performance and scalability, with both standard and lightweight variants available for English and multilingual use.
Key Features
Supports text, images, and mixed inputs (e.g., PDFs)
Multilingual embeddings for over 100 languages
Flexible output dimensions: 256, 512, 1024, or 1536 (default)
128k token context length
Multiple similarity metrics: Cosine, Dot Product, Euclidean
Batch embedding jobs and integration with popular frameworks (LangChain, LlamaIndex)
Variants include light models for faster performance
Pros & Cons
Pros
- Very long context length (128k) for processing large documents
- Broad language support (100+ languages) for global applications
- Flexible embedding dimensions allow tuning for performance vs. accuracy
- Supports multiple input types (text, images, PDFs) in a single model
- Multiple similarity metrics for different use cases
- Lightweight variants available for faster inference
Cons
- Primarily a cloud API; no local deployment options detailed
- Pricing model may be cost-prohibitive for very high volume usage
- Lightweight models trade some accuracy for speed
- Dependence on Cohere's API infrastructure
Best For
Semantic search across multilingual documentsText and image classificationClustering and topic modelingRetrieval Augmented Generation (RAG)Similarity detection and duplicate matchingRecommender systems and personalization
FAQ
What is the context length for Cohere Embed v4.0?
Embed v4.0 supports a maximum context length of 128,000 tokens.
What languages does the multilingual embed model support?
The multilingual embed model supports over 100 languages, including Chinese, Spanish, French, and many others. See the documentation for the full list.
How do I use Embed v4.0 for classification?
Embed v4.0 can generate embeddings that can be used with the Classify endpoint for any classification or analysis task. It supports text, images, and mixed inputs.