Codestral Embed
Embedding modelCard summarizedOur first embedding model specialized for code
Source: Codestral Embed launch post (no model card published)
Where to call it
- Mistral API
codestral-embed
Good for
- RAG for code completion, editing and explanation (copilots, coding agents)
- Semantic code search from natural-language or code queries
- Similarity search and duplicate detection, including license-compliance checks
Watch out for
Overview
Mistral AI · 02Codestral, front
Codestral Embed
Our first embedding model specialized for code
Typical price
$0.15 / 1M tok
Text / code input
- Dimensions
- 1536 default, up to 3072
- Max input
- 8K tokens
- Modes
- text embeddings, code retrieval
- Released
- 28 May 2025
Codestral Embed is Mistral AI's first embedding model specialized for code, aimed at retrieval over real-world code. Its dimensions are ordered by relevance, so callers can keep only the first n, and it can return int8 or binary vectors. Mistral says that even at 256 dimensions with int8 precision it beats Voyage Code 3, Cohere Embed v4.0 and OpenAI's large embedding model on its code-retrieval benchmarks. launch post; code embeddings docs
codestral-embed-2505- Typical price
- $0.15 / 1M tok
- Dimensions
- 1536 default, up to 3072 (any leading n dimensions via output_dimension)
- Max input
- 8K tokens
- Weights
- Closed
- Released
- 28 May 2025
- text embeddings
- code retrieval