Mistral has launched Mistral 3, a family of open models with strong performance and customization capabilities, supported from day one on Modal, a platform facilitating deployment without complex infrastructure management. Modal offers advanced features such as GPU memory snapshotting, which significantly reduces cold start times for these models by nearly tenfold, from around two minutes to ten seconds. The Mistral 3 suite includes multimodal models with multilingual support, available in various sizes, with the Ministral 3 being particularly optimized for Modal's serverless infrastructure. This optimization makes it appealing for companies seeking a balance of intelligence and computational efficiency. To deploy Mistral 3 models effectively, developers can use Modal's serverless GPUs and distributed file system, which streamline the integration with vLLM servers. Modal's new GPU snapshotting feature, currently in alpha, enables faster cold starts by transferring GPU memory to CPU memory upon initialization, allowing for more cost-effective and responsive deployments.