Skip to main content
You pay only for the resources you use, billed per second. Rates are in USD. Per-second rates are rounded for display. To fetch current rates from code, use Client.get_pricing() in the Python SDK.

​
Cost comparison

The rates below compare each provider’s published pay-as-you-go sandbox pricing as of August 27, 2026. See the published Modal pricing and Daytona pricing for current rates.
Modal sandboxes use gVisor containers, while Daytona sandboxes use standard Docker containers. Thunder uses Firecracker microVMs to provide strong isolation between workloads and full compatibility with Linux features.

​
Representative monthly costs

Each scenario assumes one H100 GPU per sandbox and that every sandbox runs for the listed number of hours during the month. For example, 100 sandboxes running for 200 hours each equals 20,000 total sandbox-hours.
For Thunder and Daytona, monthly cost is sandbox count × hours × (H100 rate + vCPU × vCPU rate + RAM × memory rate + billable storage × storage rate).For Modal, monthly cost is sandbox count × hours × (H100 rate + vCPU × vCPU rate + RAM × memory rate) because Sandbox storage is included.
* Modal sandboxes are preemptible by default. Non-preemptible sandboxes cost 3× more than Modal’s published rates. All Thunder sandboxes are non-preemptible.