Mistral’s new Devstral AI model was designed for coding

Mistral unveils Devstral, a 24B parameter AI coding model for efficient local deployment.

: Mistral presents Devstral, a collaboration with All Hands AI, claiming it tops other models in coding skills. Devstral is freely available under Apache 2.0, ideal for local computers, compatible with Nvidia RTX 4090 or Macs with 32GB RAM. It excels in exploring codebases and managing multiple files, and while smaller, at 24 billion parameters, Mistral plans a larger model soon. As part of Mistral's recent launches, Devstral follows Mistral Medium 3 and Le Chat Enterprise, showcasing their innovative approach to AI tools.

Mistral, founded in 2023, has introduced a new AI model named Devstral, aimed specifically at enhancing coding processes. Announced in partnership with All Hands AI, Devstral can be freely used commercially due to its open Apache 2.0 license. It's claimed to outperform competitors like Google's Gemma 3 27B and DeepSeek’s V3 on the SWE-Bench Verified, a benchmark evaluating coding proficiency. Mistral asserts that Devstral is particularly strong in navigating codebases, editing multiple files, and powering software engineering agents.

Devstral has been designed to operate on hardware like a single Nvidia RTX 4090 or a Mac with 32GB RAM, making it a suitable option for local and on-device usage. Mistral depicts this model as more efficient, able to run over code agent scaffolds such as OpenHands or SWE-Agent. These frameworks define the interface between the AI model and test cases, supporting the functional benefits of Devstral in coding environments.

Following the trend of AI-powered coding assistants, Mistral's Devstral joins recent model releases from companies like JetBrains, Google, Windsurf, and OpenAI. Despite AI models sometimes struggling with code quality and security vulnerabilities, their potential for boosting coding productivity has made them increasingly popular. A poll revealed that 76% of developers had already adopted AI tools in their development processes by the previous year.

Devstral represents Mistral's third significant product launch within a month, introduced alongside Mistral Medium 3 and Le Chat Enterprise. The latter, a corporate-focused chatbot service, demonstrates Mistral's broader ambitions to integrate AI models with third-party services, such as Gmail, Google Drive, and SharePoint. Mistral's strategic focus is on developing a larger agentic coding model, anticipated to become available soon.

Pricing for Devstral is set at $0.1 per million input tokens and $0.3 per million output tokens, with a token reflecting raw data processed by AI models. Mistral has accumulated over €1.1 billion in venture capital, with clients including BNP Paribas, AXA, and Mirakl. With its advancements, Mistral is positioning itself as a frontier lab in the AI domain, backed by investors like General Catalyst.

Sources: Mistral, TechCrunch

AIstartupsGenerative AI