Mistral has rolled out a complete AI coding stack, including Codestral 25.08, that aims to solve the key problems stopping generative AI from taking hold in enterprise software development. The integrated suite of tools is designed to create a secure, customisable, and efficient AI-native development environment.
While AI coding assistants have shown immense promise over the past year, their actual adoption inside large companies has been slow going. According to Mistral AI, this isn’t about the performance of the AI models themselves, but rather the fundamental ways these tools are built, delivered, and managed.
Many organisations – especially in heavily regulated industries like finance, defence, and healthcare – hit a wall with AI tools that are only offered as cloud services, with no way to run them on-premise or in a secure, disconnected environment.
On top of that, businesses need to tweak AI models to understand their own private codebases and internal rules, something that’s nearly impossible without access to the model’s weights or the ability to customise them.
The problems continue with fragmented systems where different AI parts don’t talk to each other properly, creating inconsistent results and a headache for operations teams. Without a single place to see what’s going on, it’s hard for companies to responsibly scale up AI use or even figure out if they’re getting a real return on their investment.
Mistral’s answer is a fully integrated system built from the ground up to meet these enterprise needs. The Mistral AI coding stack is designed to help with the entire software journey, from suggesting a line of code to handling an entire pull request on its own.
The foundation of the stack is Codestral, Mistral’s family of AI coding models fine-tuned for providing fast and accurate code completions right where developers need them. The newest version, Codestral 25.08, brings a 30% jump in the number of accepted code suggestions and is 5% better at following instructions. This model is flexible enough to be deployed in the cloud, a private virtual network, or on a company’s own servers without needing any changes to the architecture.
Of course, code completion is only useful if the model understands the project’s existing code. That’s where Codestral Embed comes in; a new embedding model created just for searching code.
Mistral reports that Codestral Embed beats leading AI coding models from OpenAI and Cohere in real-world tests for code retrieval. It offers flexible outputs to help balance performance with storage costs and can be run entirely in-house, so no sensitive data ever leaves the company’s control.
Once the right context has been found, the third piece of the puzzle, Devstral, steps in to handle complex, autonomous tasks.
Using the OpenHands agent framework, Devstral can manage tricky jobs like refactoring code across multiple files, generating tests, and writing pull requests. The open-weight Devstral Small model scores 53.6% on the SWE-Bench Verified benchmark and is efficient enough to run on a single high-end gaming GPU or a Mac with 32GB of RAM, which is perfect for self-hosted or experimental setups. For even more power, a larger Devstral Medium model is available through enterprise deals.
All of this technology is brought together neatly in Mistral Code; a native plugin for the VS Code and JetBrains IDEs. It delivers the line-by-line completions from Codestral, offers one-click automations from Devstral, and provides a built-in semantic search powered by Codestral Embed.
For the teams in IT and security, Mistral’s system gives them total operational control. The Mistral Console provides clear insight into how the AI coding tools are being used, while features like Single Sign-On and audit logs help ensure everything stays compliant with company policy.
This tight integration means a developer could start by letting Codestral draft an implementation of a new function. Then, instead of hunting through Slack or GitHub, they can ask the IDE a direct question like, “How do we handle Stripe timeouts in the checkout flow?” to see how similar problems were solved elsewhere. Finally, they could tell a Devstral agent to apply a new pattern across several different services, letting the AI do the tedious work of editing all the relevant files.
This new Mistral AI coding stack is already being put through its paces in the real world. The global consulting firm Capgemini is using it to help their teams build software faster for clients in sensitive sectors like defence and energy.
Alban Alev, VP Head of Solutioning at Capgemini France, said: “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.“
In a similar vein, the Spanish bank Abanca is using Mistral’s models in a completely self-hosted setup to meet strict European banking regulations and data privacy rules.
Mistral’s strategy reflects a wider trend in the industry. Businesses are no longer satisfied with isolated AI tools; they are looking for comprehensive, secure systems that can handle the real-world complexity of modern software development. The entire Mistral AI coding stack is available for enterprise use today.
(Image credit: Mistral)
See also: Developers adopt AI tools but question their accuracy
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.
Explore other upcoming enterprise technology events and webinars powered by TechForge here.
