French AI startup Mistral has released Devstral 2, a massive open-weights coding model boasting over 123 billion parameters. This behemoth of a model is currently available for free through the company's API and promises to outperform proprietary options in terms of cost efficiency.
According to industry benchmarking standards, Devstral 2 achieves a remarkable 72.2% score on SWE-bench Verified, a set of real-world GitHub issues designed to test AI models' ability to solve software engineering problems. This puts it firmly among the top-performing open-weights models in the market.
But what's truly innovative here is Mistral's accompanying development app called Mistral Vibe. Essentially, this is a command-line interface (CLI) that lets developers interact with Devstral models directly in their terminal. The tool can scan file structures and Git status to maintain context across an entire project, make changes across multiple files, and execute shell commands autonomously.
The release of Mistral Vibe has significant implications for the way we approach coding and collaboration. By embracing "vibe coding," a style popularized by AI researcher Andrej Karpathy, developers can describe their desired outcomes in natural language and accept AI-generated code without reviewing it closely. While this approach has its benefits – such as speed and reduced effort – it also raises important questions about the quality, understandability, and maintainability of generated code.
Mistral's CEO is confident that Devstral 2 will be able to maintain coherence across entire projects, detect failures, and retry with corrections. The company claims that this model can track framework dependencies and handle tasks like bug fixing and modernizing legacy systems at repository scale. Whether or not these claims hold water remains to be seen, but one thing is certain: the AI coding landscape has just become a lot more crowded – and competitive.
According to industry benchmarking standards, Devstral 2 achieves a remarkable 72.2% score on SWE-bench Verified, a set of real-world GitHub issues designed to test AI models' ability to solve software engineering problems. This puts it firmly among the top-performing open-weights models in the market.
But what's truly innovative here is Mistral's accompanying development app called Mistral Vibe. Essentially, this is a command-line interface (CLI) that lets developers interact with Devstral models directly in their terminal. The tool can scan file structures and Git status to maintain context across an entire project, make changes across multiple files, and execute shell commands autonomously.
The release of Mistral Vibe has significant implications for the way we approach coding and collaboration. By embracing "vibe coding," a style popularized by AI researcher Andrej Karpathy, developers can describe their desired outcomes in natural language and accept AI-generated code without reviewing it closely. While this approach has its benefits – such as speed and reduced effort – it also raises important questions about the quality, understandability, and maintainability of generated code.
Mistral's CEO is confident that Devstral 2 will be able to maintain coherence across entire projects, detect failures, and retry with corrections. The company claims that this model can track framework dependencies and handle tasks like bug fixing and modernizing legacy systems at repository scale. Whether or not these claims hold water remains to be seen, but one thing is certain: the AI coding landscape has just become a lot more crowded – and competitive.