Modal

A serverless cloud for AI workloads that runs Python functions, model endpoints, batch jobs and isolated Sandboxes for executing untrusted agent code, with per-second billing and scale-to-zero autoscaling.

Website

https://modal.com/

Who it's for

Developers who want serverless CPU/GPU compute for inference, training, batch work and agent sandboxes without idle charges.

Pricing

Starter $0/mo plus $30 free compute (3 seats, 100 containers, 10 GPU concurrency); Team $250/mo plus $100 free credits (unlimited seats, 5,000 containers, 50 GPU concurrency); Enterprise custom. Sandboxes billed at roughly 3x list compute rates.

Category

Agent operations

Last verified

2026-10-22 (current)

Confidence

high

Key facts

• Serverless and code-first: define a function with the resources it needs; Modal builds and scales it
• Sandboxes use gVisor-based isolation for untrusted code
• Supports CPU and GPU execution, arbitrary runtimes, and filesystem/memory snapshots
• Bills per second with scale-to-zero, so idle work costs nothing
• Network egress billed from October 2026

Sources

Every claim above is drawn from these; we do not publish unsourced specifications.

• https://modal.com/pricing
• https://modal.com/products/sandboxes

Notes & uncertainties

Sandboxes are non-preemptible and billed at roughly 3x standard compute. Vendor-published run-rate figures are unverified.

Something changed or out of date? Tell us — corrections are how this stays worth reading.