Codestral Embed

Mistral AI · 1 year ago · 19

Mistral released Codestral Embed, an embedding model designed for code retrieval that outperforms Voyage Code 3, Cohere Embed v4.0, and OpenAI's large embedding model on real-world code benchmarks. The model costs $0.15 per million tokens through the API and maintains better performance than competitors even when compressed to 256 dimensions with int8 precision. The embeddings enable code retrieval for AI assistants, semantic code search, duplicate detection, and repository analysis across development workflows.