About Modal
Modal is a cloud platform for running AI workloads without managing servers. It handles GPUs, containers and scaling for inference, training and sandboxes.
What it does
You write Python code and deploy it to Modal, which starts GPU containers when they are needed. The platform covers serving model endpoints, fine-tuning and reinforcement learning jobs, and isolated sandboxes for agents or untrusted code. Multi-node runs are available for larger training jobs.
Key features
- Serverless GPU containers started on demand
- OpenAI-compatible model endpoints
- Fine-tuning and reinforcement learning jobs
- Isolated sandboxes for coding agents
- Multi-node runs for large training jobs
Who it's for
- Developers deploying custom model code
- Research teams running training and evaluation jobs
- Startups that want to avoid managing servers
Pricing and good to know
Modal bills for the compute you use, and the pricing page lists current GPU rates. Startup credits are offered to some teams, so check the current offer before you begin. Cold starts can add delay when a container has not run recently, which matters for latency-sensitive endpoints.
Highlights
- Run serverless GPU containers on demand
- Serve OpenAI-compatible model endpoints
- Run fine-tuning and reinforcement learning jobs
- Isolate coding agents in sandboxes
- Scale multi-node training runs
