# Modal Modal runs CPU and GPU workloads from Python code. Serve models, train on cloud GPUs or launch batch jobs with autoscaling compute and usage logs. - Findings24: - Site: - Preços: Freemium - Categorias: Plataformas de inferência de IA, Nuvem de GPU ## O que é Modal? Modal is cloud infrastructure for developers running model inference, training, batch processing and agent workloads. Define your application, dependencies and hardware through its Python SDK; Modal packages the code into containers, runs it in the cloud and scales container capacity with demand. To start, create an account, install the Modal package and authenticate with the CLI. Define a workload and its environment, then run it remotely or deploy a function as an HTTPS endpoint. The workspace exposes logs and metrics for inspecting runs. GPU use requires a valid payment method on file, even when you plan to use included compute credits. ## O que Modal pode fazer? ### Code-defined cloud execution Describe a function's container environment and CPU/GPU requirements in Python. Modal manages the underlying capacity and container execution rather than requiring you to build a separate cluster setup. ### Inference and training workloads Serve open-weight or custom model stacks, run fine-tuning and reinforcement-learning jobs, or use single- and multi-node GPU capacity for research. ### Background jobs and web functions Fan work out across parallel containers, launch asynchronous batch jobs, or expose a function as a traffic-scaled HTTPS endpoint that can scale to zero. ### Agent sandboxes and operational visibility Create isolated environments for coding agents or other code execution, with custom images and snapshot/restore options. Inspect function, sandbox and container logs and metrics to debug the workload. ## Casos de uso ### Deploy an application-specific model service Bring your inference stack and connect it to an application through a cloud endpoint, allowing compute capacity to follow changing request volume. ### Run training and parallel data work Use GPU jobs for fine-tuning or training, and parallel functions for preprocessing, evaluations and batch processing. ### Give coding agents an execution environment Run agent-generated commands or rollout tasks inside programmatically created sandboxes, then examine outputs and logs in the workspace. ## Perguntas frequentes ### What is included in Starter? Starter has no monthly plan fee and includes $30 of compute per month, up to three workspace seats, 100 containers and 10 concurrent GPUs. Compute usage beyond the allowance and services outside its scope are billed. GPU use requires a valid payment method. ### Does the monthly allowance cover Shared Endpoints? Included compute covers Functions, Sandboxes and Notebooks using CPU, memory and GPU time. Since September 1, 2026, Shared Endpoints have been excluded from that allowance and charged per token from the first request. ### When is container compute billable? Billing includes application loading, input processing and the configurable warm period after the last input—60 seconds by default. Compute charges stop when the app scales to zero containers. CPU/memory billing uses the higher of requested or actual resources; storage, egress and token services have their own pricing. ### How do Team and Enterprise pricing differ? Team is $250 per month plus compute, with $100 of monthly compute credits, unlimited seats and higher concurrency. Enterprise uses custom pricing. Compute beyond included credits, and services outside their scope, is billed separately. ### What do I need to build on Modal? Create an account, install and authenticate the Python SDK, and define a containerized workload with its dependencies and hardware. You supply the application logic and model stack. Modal hosts the execution; its guide provides examples for inference, training, batch work and sandboxes. ## Informações rápidas - **Product**: Cloud platform for containerized CPU/GPU and AI workloads - **Primary development language**: Python - **Workload types**: Inference, training, functions, batch jobs and sandboxes - **Core free allowance**: Starter: $30 compute per month, within plan limits - **Billing distinction**: Container compute is resource-based; Shared Endpoints use separately billed tokens