With Codestral 25.08 and a complete coding stack, Mistral AI aims to accelerate the enterprise adoption of AI-driven development. The French company claims that organizations can halve their development time by using the new solution.
Mistral presents a complete toolchain consisting of four components. Together, they form an end-to-end solution for the entire development process. First, code completion, which must be fast and accurate. Second, search across the entire codebase with better performance than OpenAI and Cohere, according to Mistral AI. Thirdly, agentic workflows are of great importance, alongside integrations into existing IDEs. In this case, Mistral’s offering is compatible as a plugin with JetBrains and VS Code.
New version offers progress
The most important upgrade in this package is Codestral 25.08, a new version of Mistral’s model for generating code. Codestral has been updated to provide 30% more good (i.e., accepted) code suggestions. In addition, 10% more code is retained after modifications, and there are half as many runaway generations, i.e., AI outputs that make one error after another.
The model also works for fill-in-the-middle completion and is optimized for production environments where latency is crucial. Codestral can run in cloud, VPC, or on-premises infrastructure without requiring any infrastructure changes.
Autonomous development processes
Devstral now performs autonomous development in multiple steps simultaneously. It achieves 53.6 percent (Small variant) to 61.6 percent (Medium variant) on the SWE-Bench Verified benchmark. This outperforms Claude 3.5 and GPT-4.1-mini.
The Mistral Code platform integrates all functionalities into JetBrains and VS Code. This gives users access to inline completion, one-click automation for commit messages, and integrated semantic search functionality.
Availability
The complete Mistral coding stack is available today. Mistral AI has a roadmap for adoption in mind. Teams can start with autocomplete and semantic search functionality, then add agentic workflows later.
API keys are available for custom implementations via console.mistral.ai. Enterprise customers who want on-premise deployment or customized pilots can contact us via the request form.
Read also: Mistral launches first European AI reasoning model