Mistral AI Launches Devstral 2 Coding Model and Mistral Vibe CLI

Mistral AI has officially launched Devstral 2, a 123-billion-parameter coding model designed for enterprise software development workflows, alongside a smaller Devstral Small 2 variant and a new Mistral Vibe CLI interface for "vibe coding" workflows, according to the company's official announcement.

Devstral 2 achieves 72.2% on SWE-bench Verified, establishing it as one of the top-performing open-weight coding models while maintaining cost efficiency, the company reports. The model supports complex enterprise development tasks including multi-file edits, refactoring, and integration with agentic workflows, requiring at least four Nvidia H100 GPUs for self-hosting. Devstral Small 2, a 24-billion-parameter model, runs locally on higher-end consumer hardware like gaming-class GPUs, making it accessible for developers without enterprise infrastructure.

Both models differ in licensing: Devstral 2 uses a modified MIT license, while Devstral Small 2 employs the Apache 2.0 open-source license. This licensing strategy provides enterprises with clear commercial usage frameworks while ensuring at least one model remains fully open for local deployment and integration, as detailed in Mistral's documentation.

Initially free via API, Devstral 2 will transition to paid pricing at $0.40 per million input tokens and $2.00 per million output tokens. Devstral Small 2 will cost $0.10 per million input tokens and $0.30 per million output tokens when accessed through the API. These pricing tiers position Devstral 2 as a high-end solution and Devstral Small as a lower-cost option for latency-sensitive or local-first workflows.

Mistral frames Devstral 2 as part of a broader enterprise strategy complementing its Mistral Large foundation model, with company materials highlighting the model's focus on agentic behavior for coding, including tool calling, test generation, and codebase navigation. The company claims Devstral 2 is "up to 7x more cost-efficient than Claude Sonnet at real-world tasks," though human evaluations conducted by an independent provider showed Devstral 2 achieved a 42.8% win rate versus 28.6% loss rate against DeepSeek V3.2, while Claude Sonnet 4.5 remained significantly preferred.

The Mistral Vibe CLI, introduced alongside the models, provides a terminal-native interface for Devstral, enabling end-to-end code automation through natural language commands. As described in Mistral's documentation, Vibe features "interactive chat" with a conversational AI agent, a "built-in toolset" for file manipulation and code searching, "project-aware context" that scans file structures and Git status, and "highly configurable" settings through a simple config.toml file.

With the release of Mistral Vibe 2.0, the company has added features including custom subagents for targeted tasks, multi-choice clarifications to prevent ambiguous actions, slash-command skills for common workflows, and unified agent modes that combine tools and permissions. The CLI now ships with continuous bug fixes and improvements without requiring manual updates, as noted in Mistral's recent announcement.

Enterprise teams can access Mistral Vibe through Le Chat Pro and Team plans, with additional options for pay-as-you-go credits or bringing their own API key (BYOK). The company also offers enterprise add-ons including fine-tuning on internal languages, reinforcement learning with custom environments, and end-to-end code modernization services for migrating entire codebases to modern stacks.

For developers seeking to use the CLI with other models, Mistral Vibe supports configuration through the config.toml file to connect with models like Mistral Small, Mistral Medium, or Mistral Large, as detailed in the official documentation. This flexibility allows users to select models based on specific needs and cost considerations.

The release positions Mistral as a significant player in the open-source coding model space, directly competing with larger models like DeepSeek V3.2 and Kimi K2 while emphasizing cost efficiency and accessibility. Devstral 2's performance metrics and licensing approach represent a strategic move to provide enterprise-grade capabilities with open-source transparency, potentially influencing how developers and organizations approach AI-assisted coding workflows.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn