Modal

Platform

Serverless cloud for running Python on GPUs, billed per second

Price: Free tier, then paid
Access: Token ID and secret

About

Cloud platform where you define containers, GPUs and endpoints in Python code and pay per second of use. Used for model inference, fine-tuning, batch jobs and sandboxes. Apps are written in Python; the Go and JS/TS SDKs can call functions and run sandboxes.

What you can do with it

Deploy an open model as an autoscaling GPU inference endpoint
Fine-tune or train a model on H100 or B200 GPUs without managing servers
Run batch jobs, cron jobs and isolated sandboxes for untrusted code

Get started

Install the client with pip install modal
Run modal setup to create a token in your browser

Example

# pip install modal && modal setup, then: modal run getting_started.py
import modal

app = modal.App("getting-started")

@app.function()
def square(x: int) -> int:
    return x**2

@app.local_entrypoint()
def main(x: int = 42):
    print(f"The square of {x} is {square.remote(x)}")

Details

Hosting: Hosted service
Available in: Worldwide
Official SDKs: Python, Go, JavaScript/TypeScript
MCP server: None