Mistral is entering the competitive world of AI-powered developer tools with the launch of Mistral Code — a coding assistant built for real-world engineering teams and enterprise developers.
Highlights
- Mistral Code enters private beta with support for VS Code and JetBrains IDEs, offering integrated AI-powered coding assistance tailored for real-world developers.
- Powered by a suite of proprietary models — including Codestral, Devstral, and Codestral Embed — optimized for code generation, search, reasoning, and multi-step workflows.
- Enterprise-first focus with deployment flexibility (cloud, private infra, or air-gapped), fine-tuning on private codebases, and robust team admin controls.
- Already in use by large organizations such as Capgemini, Abanca, and SNCF, showing early traction in secure, production-grade environments.
- Codestral model boasts 22B parameters and a 32k-token context window, supporting over 80 programming languages with strong performance on HumanEval, RepoBench, and others.
- Deep integration with developer workflows via LangChain, LlamaIndex, and Google Cloud Vertex AI, ensuring minimal friction and high adaptability.
- Open-source aligned — Mistral commits to contributing back to the Continue project and has made Codestral accessible via Hugging Face and its own API portal.
- Part of a growing AI ecosystem from Mistral, which also includes Le Chat Enterprise and integrations with popular platforms like Gmail and SharePoint.
- This launch strengthens Mistral’s ambition to become Europe’s go-to provider for transparent, enterprise-grade, and developer-focused AI solutions.
Now available in private beta for JetBrains IDEs and Microsoft’s VS Code, Mistral Code is designed to deliver powerful, integrated coding support within familiar environments.
With a strong focus on customization, on-premise deployment, and multilingual code support, Mistral positions its tool as a practical alternative to GitHub Copilot, Cursor by Anysphere, and other incumbents.
A Stack of Specialized Models
Mistral Code is built on a fork of the open-source Continue project and incorporates multiple proprietary AI models developed by Mistral:
- Codestral: The primary model powering autocomplete and code generation
- Codestral Embed: Optimized for semantic code search and retrieval
- Devstral: Designed for complex, multi-step tasks and agentic workflows
- Mistral Medium: Provides chat-style developer assistance
Built for Enterprises — Not Just Indie Devs
While many AI coding tools cater to individual developers, Mistral is targeting enterprises from the outset.
- Flexible deployment options: Cloud-based, dedicated reserved infrastructure, or air-gapped local GPU support
- Fine-tuning capabilities: Model customization on private repositories
- Admin tools: Role management, usage analytics, and seat controls for team environments
Mistral describes the platform as “best-in-class AI tooling for enterprise developers,” aiming to serve teams with complex infrastructure and privacy requirements.Real-World Adoption Already Underway
Mistral Code is already being used in production environments by several major organizations, including,
- Capgemini
- Abanca
- SNCF (France’s national railway operator)
This early adoption signals strong interest from sectors that demand security, traceability, and integration into complex CI/CD pipelines.
Codestral’s Capabilities
At the core of Mistral Code is Codestral, an open-weight generative model trained on over 80 programming languages — from Python, Java, and JavaScript to more specialized languages like Swift, Rust, and Fortran.
Specs:
- 22 billion parameters
- 32,000-token context window for handling large codebases
- Strong performance on benchmarks like HumanEval, MBPP, RepoBench, and CruxEval
Despite its relatively compact size, Codestral is optimized for low latency and excels in long-range code completion.
Integrated, Not Isolated
Codestral is built to integrate deeply into existing workflows. It supports:
- Popular IDEs like VS Code and JetBrains
- Developer frameworks such as LangChain and LlamaIndex, enabling agent-based applications
- Google Cloud’s Vertex AI (via recent partnership), offering broader deployment possibilities
Accessibility, Community, and Open Source Commitment
Mistral has made Codestral available on Hugging Face and via its own API platform, enabling developers to experiment, test, and provide feedback. The assistant is also accessible through Le Chat, Mistral’s conversational AI platform.
In a nod to its open-source roots, Mistral has committed to contributing improvements from Mistral Code back to the Continue project — reinforcing the company’s dual focus on innovation and transparency.
Mistral’s Growing AI Portfolio
Mistral Code is part of a larger suite of AI tools from the Paris-based startup, which launched in 2023 and has already raised over €1.1 billion (approx. $1.24 billion USD).
Other offerings include Le Chat Enterprise, an AI assistant designed for integration with platforms like Gmail, Google Drive, and SharePoint.
