Modal Freemium

Modal is a serverless cloud for AI workloads. You decorate a Python function, run one command, and it becomes a scalable GPU endpoint that bills per second and scales to zero when idle.

What is Modal?

Modal, founded by Erik Bernhardsson and Akshat Bubna, targets the gap between “my laptop” and “a Kubernetes cluster.” You write normal Python, add a decorator, and Modal handles containers, GPUs, autoscaling and HTTPS endpoints. There is no Dockerfile to babysit and no YAML.

Modal features that matter

• GPU on demand: request A100, H100 or other accelerators per function.
• Scale to zero: idle apps cost nothing, with request-driven cold starts.
• Python-native: define images, secrets and volumes in the same file as your code.
• Web endpoints: turn a function into an HTTP API with one decorator.
• Volumes and queues: persistent storage plus async job pipelines.

Product highlights

Developers like it because deployment stops being a project. Fine-tuning a model or hosting a diffusion endpoint is a few dozen lines, not a DevOps sprint. Billing is per second, so a job that runs 40 seconds doesn’t rent a GPU for an hour.

Watch the cold starts. Scale-to-zero is cheap but a cold GPU boot can take tens of seconds, so latency-sensitive apps often keep a warm pool — which costs more.

Who should use Modal

Great for ML engineers and backend devs who know Python and want GPU work without infra chores. Less appealing if you are locked into another cloud’s ecosystem or need Windows or non-Python runtimes.

Common use cases

• Hosting an image or video generation model behind an API.
• Batch inference over millions of records.
• Scheduled jobs like nightly embeddings refreshes.
• Fine-tuning runs without reserving a cluster.
• Prototyping a GPU app in an afternoon.

Pricing and limits

Modal charges per CPU/GPU-second with a monthly free compute grant aimed at hobby projects. Reserve capacity is available for latency-sensitive work. Exact GPU rates and free-tier limits are listed on their pricing page.

Quick Summary

• Modal runs Python functions on cloud GPUs with no Docker or Kubernetes.
• Billing is per second, and idle apps scale to zero.
• Free monthly compute grant for small projects.
• Cold starts are the main latency trade-off.
• Best for Python-first teams doing AI/GPU work.