Mistral AI introduces Code programming assistant

Mistral AI SAS today debuted Mistral Code, a tool that uses four of its artificial intelligence models to help developers write code faster.

Paris-based Mistral AI raised $640 million last year at a $6 billion valuation. Alongside AI programming tools, it provides general-purpose large language models that compete with OpenAI’s algorithms. The company also offers tools that enterprises can use to build AI agents.

Mistral Code is available as an extension for several popular code editors. Developers can access the tool directly from a supported editor’s interface without opening a new tab, which streamlines day-to-day use. The software generates programming advice in response to natural language requests entered by the user.

Under the hood, Mistral Code is based on an open-source project called Continue. It’s a collection of ready-to-use building blocks for creating AI coding assistants. Mistral AI expanded the project’s feature set with capabilities that allow companies to manage their developers’ Mistral Code accounts and monitor usage.

The company also combined Continue with four AI models. Each processes a different subset of the requests that users send to Mistral Code.

The first model, Codestral Embed, powers Mistral Code’s search features. Developers can run queries to find specific components or embedded documentation in an application’s code base. Codestral Embed makes application code searchable by turning it into embeddings, mathematical structures that AI models use to hold information.

Mistral Code also provides autocomplete features. When a developer starts typing a line of code that appears frequently in an application, the tool can generate the rest. This feature is powered by an open-source model called Codestral that Mistral released last May.

More complicated programming tasks are related to Devstral, another open-source coding model. Mistral released it last month in partnership with a startup called All Hands AI. According to the company, Devstral outperforms OpenAI’s GPT-4.1-mini by more than 20% on a popular AI programming benchmark. A fourth model called Mistral Medium powers Mistral Code’s chatbot interface.

Organizations with advanced requirements can customize the algorithms that power the tool. “Customers can fine-tune or post-train the underlying models on private repositories or distill lightweight variants—capabilities that simply don’t exist in closed copilots tied to proprietary APIs,” Mistral staffers wrote in a blog post today.

The launch of Mistral Code will create more competition for OpenAI. Last month, the ChatGPT developer acquired AI coding assistant startup Windsurf Inc. in a deal reportedly worth $3 billion. Windsurf initially offered its software as an extension for popular code editors before launching its own custom, AI-optimized code editor. Mistral may seek to take a similar approach in the future to more directly compete with OpenAI.

Image: Mistral AI