Mistral AI Codestral 25.08 and Enterprise Coding Stack Release

Mistral AI has announced the release of Codestral 25.08 and a complete integrated coding stack designed for enterprise-grade software development. This system combines high-precision code completion, codebase-scale semantic retrieval, and autonomous agentic workflows to reduce development, review, and testing time by up to 50% while supporting strict deployment requirements like VPC and on-premises environments.

Codestral 25.08: High-Fidelity Code Completion

Codestral 25.08 is the latest update to Mistral's family of code generation models, specifically optimized for fill-in-the-middle (FIM) completion. Validated in live IDE usage across production codebases, the 25.08 version introduces the following performance improvements:

  • Accepted Completions: 30% increase in accepted completions.
  • Code Retention: 10% increase in retained code after suggestions.
  • Stability: 50% fewer runaway generations, increasing confidence in longer edits.
  • Chat Capabilities: A 5% increase in instruction following (IF eval v8) and a 5% increase in average MultiplE code abilities.

The model is deployable across cloud, VPC, or on-premises environments without requiring architectural changes.

Codestral Embed: Semantic Retrieval for Large Codebases

Codestral Embed provides a specialized embedding layer designed for code rather than general text, enabling high-recall, low-latency search across massive monorepos and poly-repos.

Key technical advantages include:

  • Performance: Outperforms leading embedding models from OpenAI and Cohere in real-world code retrieval benchmarks.
  • Efficiency: Offers configurable dimensions (e.g., 256-dim, INT8) to balance retrieval quality with storage efficiency.
  • Privacy: Supports private deployment, allowing all embedding inference and index storage to run within enterprise infrastructure to prevent data leakage.

Devstral: Autonomous Agentic Workflows

Devstral, powered by the OpenHands agent scaffold, enables multi-step engineering tasks such as cross-file refactors, test generation, and PR authoring.

Model Performance and Availability

Devstral is available in two primary sizes to balance performance and accessibility:

  • Devstral Small (24B): An open-weight model (Apache-2.0) that can run on a single Nvidia RTX 4090 or a Mac with 32 GB RAM. It is suitable for self-hosted, air-gapped, or experimental workflows and can be fine-tuned on proprietary code.
  • Devstral Medium: Available via API and enterprise partnerships for advanced planning and code understanding. This version supports post-training and fine-tuning for enterprise clients.

On the SWE-Bench Verified benchmark, Devstral Small 1.1 scores 53.6% and Devstral Medium reaches 61.6%, outperforming Claude 3.5 and GPT-4.1-mini.

Mistral Code: IDE Integration and Operational Control

All components of the stack are integrated into the Mistral Code plugin for JetBrains and VS Code. This plugin provides a unified interface for inline completions, one-click task automations (e.g., "Fix function" or "Add docstring"), and integrated semantic search.

Enterprise Governance and Security

To address the blockers typical of SaaS-only tools, Mistral Code includes:

  • Deployment Flexibility: Support for cloud, self-managed VPC, and fully on-premises deployments (the latter reaching General Availability in Q3).
  • Security Controls: Integration with SSO, audit logging, and usage controls.
  • Privacy: No mandatory telemetry and no external API calls for inference or search.
  • Observability: Usage metrics, including suggestion acceptance rates and agent adoption, are tracked via the Mistral Console.

Enterprise Adoption and Use Cases

Leading organizations are utilizing the stack to manage complex, regulated environments:

  • Capgemini: Uses the stack to accelerate development for clients in defense, telecom, and energy while maintaining compliance.
  • Abanca: Employs a fully self-hosted deployment to meet European banking regulations regarding data residency and network isolation.
  • SNCF: Utilizes agentic workflows to modernize legacy Java systems with human-in-the-loop oversight.

"Leveraging Mistral’s Codestral has been a game changer in the adoption of private coding assistant for our client projects in regulated industries. We have evolved from basic support for some development activities to systematic value for our development teams". — Alban Alev, VP head of Solutioning at Capgemini France.

Sources

Related

  • Dispatch
    Mistral AI Devstral Models Release

    Mistral AI has released Devstral Medium and Devstral Small 1.1, providing state-of-the-art agentic coding capabilities with a focus on generalization across different prompts and agentic scaffolds.

  • Dispatch
    Codestral 25.01 release notes / what's new

    Mistral AI has released Codestral 25.01, a coding model featuring a more efficient architecture and improved tokenizer that generates code approximately twice as fast as its predecessor.

  • Dispatch
    Mistral Code Release Notes

    Mistral AI has launched Mistral Code, a vertically integrated AI coding assistant for enterprises that combines frontier models with an in-IDE assistant and flexible deployment options.

  • Dispatch
    Mistral AI Devstral Release

    Mistral AI has released Devstral, an agentic LLM for software engineering that outperforms open-source models on the SWE-Bench Verified benchmark and is available under the Apache 2.0 license.

  • Dispatch
    Codestral Release Notes

    Mistral AI has released Codestral, a 22B open-weight generative AI model specifically designed for code generation across 80+ programming languages.