# Modal

Serverless compute platform whose Sandboxes run untrusted or agent-written code in gVisor or VM containers, billed per second for the larger of requested and used CPU and memory.

- Operated by: Modal Labs, Inc.
- Status: Active [1]
- Website: https://modal.com
- Categories: Agent sandboxes
- Checked: 31 Oct 2026

## Modal Sandboxes (Agent sandboxes)

- Price: $0.142 /core (2 vCPU)-h (+ $0.024 per GB-hour)
- 1-hour session: ≥ $0.238 (2 vCPU / 4 GB · rank 21 of 26)
- Fleet month: ≥ $2,094 (50 × 8 h × 22 days)
- Billed on: Running time (per second)
- Isolation: gVisor (sessions up to 24 h)

### Plans

- Starter: Free; $30/month free compute, 3 seats, 100 containers + 10 GPU concurrency, 1 TiB/month egress [1]
- Team: $250 /month; $100/month free compute, unlimited seats, 5,000 containers + 50 GPU concurrency, 10 TiB/month egress, static IP proxy [1]
- Enterprise: unknown (Custom pricing with volume-based discounts); Custom concurrency, 100 TiB/month egress, SSO, audit logs, HIPAA [1]

### Billing

- Payment: Metered: usage billed afterwards [2]
- Plan minimum: None [2]
- Billing increment: Per second [2]
- Minimum charge: unknown (No minimum billed duration on the pricing page or the Sandbox docs) [2]
- Per-job fee: None [2]

### Runtime

- Billed on: Running time [2]
- Isolation: gVisor [2]
- Max session: 24 h [2]
- Concurrent sandboxes: 100 [2]
- Largest vCPU: unknown (Docs say a maximum CPU request is enforced at creation but do not state it; the default soft CPU limit is 16.125 physical cores) [2]
- Largest memory: unknown (Docs say a maximum memory request is enforced at creation but do not state it) [2]

### Lifecycle

- Survives a stop: Nothing (ephemeral) [4]
- Snapshots: Yes [4]
- Pause and resume: No [4]
- Fork: Disk copy [4]
- Compute billed while stopped: unknown (No paused or stopped state exists; a running idle Sandbox still bills max(request, actual), at least 0.125 cores and 128 MiB) [4]
- Auto-stop: after 5 min [4]
- Start time (vendor claim): 1 s [4]

### Capabilities

- GPU: Yes [5]
- Desktop: None [5]
- Coding agents installed: None [5]
- Agent prompt API: No [5]
- SSH: Yes [5]
- Docker inside: Yes [5]
- SDKs: Python, TypeScript, Go [5]
- SOC 2: Yes [5]

### Storage and network

- Disk included: 512 GB [1]
- Extra disk: Free [1]
- Snapshot storage: unknown (Snapshots are stored as Images; no storage price for Images is published) [1]
- Egress: $0.04 per GB [1]

### Platform

- Regions: us, eu, ap, uk, ca, me, sa, af, mx [3]
- Your cloud (BYOC): unknown (No bring-your-own-cloud offer on the pricing, Sandbox or security pages) [3]
- Self-host: No [3]

### Free allowance

- Minutes a month: None [1]
- Credit a month: $30 [1]
- One-time credit: None [1]

### Prices

| Resource | Price | Per hour |
|---|---|---|
| per core (2 vCPU) | $0.0000394 /s [1] | $0.142 |
| per GB of memory | $0.00000667 /s [1] | $0.024 |

- The CPU rate is per physical core, which Modal equates to 2 vCPU; the memory rate is per GiB.
- Sandbox rates are the non-preemptible rates (3x the Function list price); billing is max(request, actual) per second.
- Pinning a region multiplies the price by 1.15x (broad, e.g. us) or 1.75x (narrow, e.g. us-west).
- GPU Sandboxes use the standard GPU rates per second: B300 $0.001972, B200 $0.001736, H200 $0.001261, H100 $0.001097, RTX PRO 6000 $0.000842, A100 80 GB $0.000694, A100 40 GB $0.000583, L40S $0.000542, A10 $0.000306, L4 $0.000222, T4 $0.000164; GPU Sandboxes can be preempted.
- Volumes cost $0.09 per GiB a month with 1 TiB a month free.
- Filesystem snapshots are kept 30 days by default (configurable); memory snapshots are alpha and expire after 7 days.
- The under-one-second start is Modal's own claim.

## Sources

[1] https://modal.com/pricing (checked 31 Oct 2026)
[2] https://modal.com/products/sandboxes (checked 31 Oct 2026)
[3] https://modal.com/docs/guide/region-selection (checked 31 Oct 2026)
[4] https://modal.com/docs/guide/sandboxes (checked 31 Oct 2026)
[5] https://modal.com/docs/guide/security (checked 31 Oct 2026)