TechPrufer
Get started
Mistral AI
Mistral AI logoMistral AI · Feature trail

Codestral Embeddings

1 post · May 2024 – Jul 2025

What this trail covers

Mistral AI continues to evolve its coding stack with Codestral 25.01, enhancing code completion accuracy and chat capabilities. The model features a more efficient architecture and an improved tokenizer, generating and completing code about 2 times faster, making it the leader for coding in its weight class and SOTA for FIM use cases. The integrated system now includes advanced codebase-scale semantic retrieval via Codestral Embed, which offers flexible embedding dimensions and private deployment.

All Mistral AI trails
  1. Announcing Codestral 25.08 and the Complete Mistral Coding Stack for Enterprise | Mistral AIThis post announces Codestral 25.08, detailing a +30% increase in accepted completions, +10% more retained code, and 50% fewer runaway generations. It also highlights improvements in chat mode instruction following (+5% on IF eval v8) and code abilities (+5% in average MultiplE). The post emphasizes the integration of Codestral Embed for codebase-scale semantic retrieval, noting its superior performance over OpenAI and Cohere on code retrieval benchmarks and flexible embedding outputs. Devstral is presented as the agentic workflow component, with Devstral Small 1.1 scoring 53.6% and Devstral Medium reaching 61.6% on SWE-Bench Verified. The Mistral Code IDE plugin is detailed for its integration of these components and enterprise deployment features.
  2. Codestral Embed | Mistral AIIntroduces Codestral Embed, a specialized embedding model for code retrieval. It demonstrates superior performance compared to leading competitors (Voyage Code 3, Cohere Embed v4.0, OpenAI's large embedding model). The model offers configurable embedding dimensions and precision (e.g., 256 dimensions with int8 precision), allowing trade-offs between retrieval quality and storage costs. Use cases include RAG for code assistants, semantic code search, similarity search/duplicate detection, and code analytics. It is available via API and batch API, with options for on-prem deployment. Recommends chunking parameters for retrieval use cases.
  3. Codestral 25.01 | Mistral AICodestral 25.01 features an improved architecture and tokenizer, resulting in 2x faster code generation and completion, and achieving SOTA for FIM use cases. It also introduces advanced codebase-scale semantic retrieval via Codestral Embed with flexible embedding dimensions and private deployment options.
  4. Codestral | Mistral AIIntroduces Codestral, a 22B open-weight generative AI model for code generation, fluent in 80+ programming languages. It offers fill-in-the-middle capabilities, improved performance and latency compared to previous models, and a 32k context window outperforming competitors in RepoBench. Benchmarks are provided for Python, SQL, and multiple other languages. The post details deployment options via a dedicated endpoint, la Plateforme API, and self-deployment, as well as integrations with LlamaIndex, LangChain, Continue.dev, and Tabnine.