High-performance, collaborative GPU notebooks in the cloud
Run or deploy machine learning models, massively parallel compute jobs, task queues, web apps, and much more, without your own infrastructure.
Modal Notebooks provide sub-second cold starts and instant autoscaling, allowing data scientists to quickly test model changes, run inference, and fine-tune models without waiting for infrastructure provisioning. The serverless model eliminates capacity planning overhead.
Modal Notebooks provide instant access to B200s, H100s, A100s, A10Gs, and other GPUs globally distributed across clouds. Users can select specific hardware per function and scale elastically, paying per-second with no reserved capacity requirements.
Modal Notebooks enable users to write Python code that automatically scales across GPUs and CPUs in the cloud. Users can execute resource-heavy workloads like model training, batch inference, and data processing with instant autoscaling from 0 to 1000+ GPUs, paying only for actual compute time used.
Modal Sandboxes (integrated with Notebooks) provide secure, ephemeral execution environments using gVisor containerization. This enables safe execution of coding agents, AI-generated code, and user submissions with full isolation and automatic cleanup.
Modal Notebooks support dynamic batching, job queues, and parallel processing primitives that allow users to spawn millions of parallel jobs in seconds. Users can process large datasets with simple Python code while Modal handles orchestration and scaling.