Enhancing AI Coding Agents with Mistral's New Devstral Models

Mistral AI has recently unveiled two advanced AI models, Devstral Small 1.1 and Devstral Medium, designed to push the boundaries of agentic coding assistance. Developed in collaboration with All Hands AI, these models aim to provide versatile, efficient, and powerful tools for software developers and organizations seeking to automate and augment their coding workflows.

Devstral Small 1.1 continues the legacy of the original Devstral Small with a focus on improved generalization across diverse prompts and programming scenarios. Despite retaining the same architecture and 24 billion parameters, this version demonstrates notable enhancements in coding performance, achieving a leading score of 53.6% on the SWE-Bench Verified benchmark without test-time scaling. Released under the permissive Apache 2.0 license, Devstral Small 1.1 invites open-source communities to explore, customize, and accelerate autonomous software development.

Moreover, Devstral Small 1.1 integrates seamlessly with OpenHands and supports multiple function invocation formats, including Mistral's proprietary function calling and XML. This flexibility allows it to adapt to a broad spectrum of coding environments and agent frameworks, maximizing its utility for various application needs.

Devstral Medium builds upon these foundations to deliver even stronger coding capabilities, registering a 61.6% score on SWE-Bench Verified. Accessible via Mistral AI's public API, it balances high performance with affordability—offering competitive pricing that undercuts comparable models like Gemini 2.5 Pro and GPT 4.1. This makes Devstral Medium an attractive choice for enterprises and developers prioritizing both efficiency and cost-effectiveness.

For organizations requiring enhanced control and privacy, Devstral Medium supports deployment on private infrastructure, allowing tailored configurations and data security compliance. Additionally, custom finetuning options empower enterprises to optimize the model for unique domains or specialized tasks, ensuring alignment with specific business requirements.

Both models follow a transparent pricing model aligned with Mistral's earlier offerings: Devstral Small 1.1 at $0.1 per million input tokens and $0.3 per million output tokens, and Devstral Medium at $0.4 per million input tokens plus $2 per million output tokens.

By releasing Devstral Small 1.1 openly and providing accessible API integration for Devstral Medium, Mistral AI emphasizes community engagement and practical accessibility. This approach supports developers and enterprises alike in leveraging cutting-edge AI to streamline content generation, code automation, and software development processes.

For marketers, developers, and SEO specialists interested in integrating AI-powered content and code automation, exploring models like Devstral offers a pathway to accelerate production workflows, improve consistency, and reduce manual burdens while maintaining adaptability and control.