Overview
mistral-embed is Mistral AI's general-purpose embedding model that transforms text into 1024-dimensional vector representations, capturing semantic meaning for various NLP tasks.
Technical Specifications
- Dimensions: 1024
- Normalization: Norm 1 vectors (cosine similarity, dot product, and Euclidean distance are equivalent)
- Processing: Batch processing support for improved efficiency
- Input: Any text length
Companion Model: Codestral-Embed
Mistral also offers codestral-embed for code-specific use cases:
- Dimensions: Up to 3072 (configurable via output_dimension)
- Purpose: Code search, repository analysis, coding assistants
- Use Cases: Semantic code search, duplicate detection, code analytics
Use Cases
- Retrieval Systems: Power RAG pipelines with semantic retrieval
- Clustering: Group similar documents or code snippets
- Classification: Categorize text at scale
- Semantic Search: Find conceptually similar content
- Duplicate Detection: Identify similar or duplicate content
- Code Search: Navigate codebases semantically
Distance Metrics
Due to norm 1 normalization, all these metrics are equivalent:
- Cosine similarity
- Dot product
- Euclidean distance
API Integration
Available through Mistral AI's Embeddings API with support for:
- Batch processing
- Multiple text inputs
- Configurable output dimensions (codestral-embed)
Integration Support
- LangChain
- LlamaIndex
- Qdrant
- Elasticsearch
- Custom implementations
