Mistral AI introduces Devastral 2, a next-generation coding model family for software engineering agents, combined with Mistral Vibe CLI, an open source command line coding assistant that runs inside a terminal or IDE that supports agent communication protocols.
Devastral 2 and Devastral Small 2, model sizes, references and benchmarks
Devstral 2 is a 123B parameter dense transformer with a 256K token context window. It reaches 72.2 percent on SWE-Bench verified, placing it among the strongest open weight models for software engineering tasks. The model is released as open source under the Modified MIT License and is currently free for use through the Mistral API.
Devstral Small 2 is a 24B parameter model with the same 256K reference window. It scores 68.0 percent on SWE-Bench Verified and falls into the category of models that are up to 5 times larger in parameter counts. It is released under the Apache 2.0 license, which is a standard permissive license for production use.
Both models are described as open source and permissively licensed and are positioned as state-of-the-art coding models for agentic workloads. Mistral reports that Devstral 2 is 7x more cost-effective than Cloud Sonnet on real-world coding tasks at similar quality, which is critical for sustained agent workloads.
In terms of model size relative to the Frontier system, Devestral 2 and Devestral Small 2 are 5 times and 28 times smaller than DeepSeek V3.2, and 8 times and 41 times smaller than KmK2.
Built for production grade coding workflows
Devstral 2 is designed for software engineering agents who need to locate repositories, track dependencies, and orchestrate edits across multiple files while maintaining architecture level context. The model can detect failures, retry with fixes, and support tasks such as bug fixing at repository scale or modernizing legacy systems.
Mistral says that Devastral 2 can be fine-tuned to favor specific programming languages or to be optimized for very large enterprise codebases. Devastral Small 2 brings the same design goals to a smaller footprint that is suitable for local deployment, tighter feedback loops, and a completely private runtime. It also supports image input and can run multimodal agents that must reason on both code and visual artifacts such as diagrams or screenshots.
Human evaluation against DeepSeek v3.2 and Cloud Sonnet 4.5
To test real-world coding behavior, Mistral evaluated Devstral 2 against DeepSeek v3.2 and Cloud Sonnet 4.5 using tasks created through the Kline Agent tool. In these human evaluations, Devstral 2 shows a clear advantage over DeepSeek V3.2 with a 42.8 percent win rate versus a 28.6 percent loss rate.
Mistral Vibe CLI, a terminal native coding agent
Mistral Vibe CLI is an open source command line coding assistant written in Python and powered by the Devastral model. It detects, modifies, and executes changes to the codebase using natural language in the terminal or inside an IDE that supports agent communication protocols like Jade where it is available as an extension. The project is released on GitHub under the Apache 2.0 license.
Vibe CLI provides a chat style interface on top of many key tools,
- Project aware context, it scans the file structure and Git state to create a working view of the repository.
- Smart Reference, it supports
@autocomplete for files,!For shell commands and slash commands for configuration changes. - Multi-file orchestration, causes this to coordinate changes at the architecture level and reduce pull request cycle times, not just on the active buffer, but on the entire codebase.
- Persistent history, autocompletion and themes tuned for daily use in the terminal.
Developers configure Vibe CLI through a simple config.toml File where they can point to Devstral2 via the Mistral API or other local or remote models. The tool supports programmatic runs, auto approval toggles for tool execution, and granular permissions to require confirmation for risky operations in sensitive repositories.
key takeaways
- Devstral 2 is a 123B parameter dense coding model with 256K references, it reaches 72.2 percent on SWE bench verified and is released as open weights under the modified MIT license.
- Devstral Small 2 has 256B parameters with the same 24K reference, scores 68.0 percent on SWE Bench Verified and uses the Apache 2.0 license for easy production adoption.
- Both Devastral models are optimized for agentive coding workloads, designed to explore full repositories, track dependencies, and enforce multi-file edits with failure detection and retries.
- Mistral Vibe CLI is an open source Python based terminal native coding agent that connects to Devstral, it provides project aware context, smart context and multi file orchestration through a chat style interface in a terminal or IDE that supports agent communication protocols.
check it out Full technical details hereFeel free to check us out GitHub page for tutorials, code, and notebooksAlso, feel free to follow us Twitter And don’t forget to join us 100k+ ml subreddit and subscribe our newsletterwait! Are you on Telegram? Now you can also connect with us on Telegram.
Asif Razzaq Marktechpost Media Inc. Is the CEO of. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. Their most recent endeavor is the launch of MarketTechPost, an Artificial Intelligence media platform, known for its in-depth coverage of Machine Learning and Deep Learning news that is technically robust and easily understood by a wide audience. The platform boasts of over 2 million monthly views, which shows its popularity among the audience.
🙌 Follow MarketTechPost: Add us as a favorite source on Google.
