AI code-execution sandbox

Modal

Modal is a Python-first serverless compute platform whose Sandboxes offer gVisor isolation, sub-second cold start and — uniquely — GPUs (T4 through B200) inside the sandbox.

Last verified 2026-07-26

Modal is the ML/Python heavyweight of the category and the standout when you need GPU inside the sandbox. It uses gVisor isolation, sub-second cold starts, and integrates natively with the OpenAI Agents SDK. Reportedly ~$300M annualized revenue.

Key specs

IsolationgVisor
Cold start800ms
Price / vCPU-hr$0.14
Pricing modelPer-second usage
GPUT4 → B200, incl. H100
Session limitConfigurable
PersistenceVolumes; ephemeral compute
LanguagesPython (first-class), others via containers
MCP support
IntegrationsOpenAI Agents SDK
DeploymentManaged cloud
Free tier$30/mo free credits
Core price~$0.14/core-hr
Revenue~$300M ARR (reported)

Strengths

  • GPU inside the sandbox
  • Sub-second cold start
  • Excellent Python ergonomics
  • Native OpenAI Agents SDK

Trade-offs

  • Python-centric
  • Higher per-core price

Best for: Python/ML workloads and agents that need GPU inside the sandbox.

Head-to-head

Alternatives to Modal

Sources: Modal pricing ↗ · Modal Sandboxes docs ↗

Get pricing for Modal

Tell us what you're building. We'll route you to the right sandbox teams for pricing and a demo — no spam.