Mistral AI has launched Vibe, a command-line interface (CLI) coding agent, alongside the public preview of its Mistral Medium 3.5 model. Vibe facilitates software development tasks such as code writing, refactoring, and test generation, with capabilities for cloud-based background execution and parallel operation of multiple agents. It also integrates with GitHub to automate pull requests and user notifications. The Mistral Medium 3.5 model, featuring 128 billion parameters, is Mistral AI's first flagship merged model. It supports a 256K context window and excels in instruction following, reasoning, and coding tasks. Trained with a visual encoder, it achieved a 77.6% score on the SWE-Bench validation set, outperforming previous models like Devstral 2. This model is now the default for Vibe and Le Chat, with configurable inference effort per request.