ModalPaid
Overview
Modal is per-second GPU cloud: decorate a Python function and it deploys, attaching an A100 or H100 when needed and releasing it the moment the job ends, which suits inference bursts, batch generation and fine-tuning without Kubernetes.
Cold starts are optimised to a few seconds. You pay for what you use, so it is efficient for spiky work; long-running services cost more than a plain VM, and deploying means writing code.
What it does
- Python functions deploy with on-demand GPU attachment
- Per-second billing with automatic release
- Optimised cold starts, usable in seconds
- Scheduled jobs, queues and persistent volumes
Best forBatch image and video generation, Model fine-tuning and inference services, Spiky compute demand
FAQ
Does Modal have a free plan?
Usage-based pricing per second of compute; free credits on signup.
When should I use Modal?
Typical uses: Batch image and video generation; Model fine-tuning and inference services. Key capability: Python functions deploy with on-demand GPU attachment.
What are the downsides of Modal?
Things to watch: Code-based deployment; always-on services cost more than VMs|Latency from mainland China can be poor.