title: "Codestral Mamba | Mistral AI"

Mistral AI has released Codestral Mamba, a new model that explores an alternative to the Transformer architecture. Based on Mamba, it offers linear-time inference and the theoretical ability to model infinite-length sequences, making it highly efficient for tasks with long inputs. The model is specifically trained for advanced coding and reasoning, demonstrating performance on par with leading transformer-based models. Codestral Mamba is available for free use and modification under an Apache 2.0 license, with weights accessible on HuggingFace and deployment options via the mistral-inference SDK.