Modal
thecontinuitylayer.com/tools/gpu-hosting/modalPython developers who want to serve an open model as code, paying only while it runs
Serverless GPU · Modal Labs
Modal is a serverless platform where you define a function or web endpoint in Python, attach a GPU type, and Modal runs it in the cloud. It starts containers on demand and scales them back to zero when idle, billing per second of use.
For open models, you typically wrap vLLM or a similar server in a Modal app and deploy it as an endpoint. Cold starts and keeping containers warm are the main things to plan for, since warm idle time is billed. Choosing a specific region adds a price multiplier.
Specs
| Price per hour | L4 $0.000222/sec (about $0.80/hr); A100 80GB $0.000694/sec (about $2.50/hr); H100 $0.001097/sec (about $3.95/hr) |
|---|---|
| GPUs | Serverless NVIDIA GPUs: T4, L4, A10, L40S, A100, H100, H200, B200 and B300 |
| Billing | Per second, scales to zero |
| L4 | $0.000222/sec (about $0.80/hr) |
| L40S | $0.000542/sec (about $1.95/hr) |
| A100 80GB | $0.000694/sec (about $2.50/hr) |
| H100 | $0.001097/sec (about $3.95/hr) |
| B200 | $0.001736/sec (about $6.25/hr) |
| Free credits | $30 of free compute per month on the Starter plan |
Good for
- Spiky or low-traffic inference endpoints
- Developers who prefer code over consoles
- Batch jobs that fan out across many GPUs
Not for
- Steady 24/7 load, where a rented GPU is cheaper
- People who want a plain SSH machine
See also
- Nebius AI Cloud · Teams that want a full-featured AI cloud with self-service access to recent NVIDIA GPUs
- Runpod · Builders who want a cheap GPU by the hour or a serverless endpoint for an open model
- Vast.ai · Cost-conscious tinkerers and researchers who are comfortable picking a host themselves
We last reviewed Modal on 31 October 2026. Prices and specs change; check modal.com before you buy or sign up. Figures we could not confirm on the vendor's own page read “not published” or “varies”.
Vendor source: https://modal.com/pricing