# Superserve

Firecracker microVMs for long-running agents, with full-state pause, snapshots and forks.

Compare it with others: [/providers/compare?p=superserve](/providers/compare?p=superserve).

## Price

| Size | Asked for | Per hour | Per month | Runs as | Plan |
| --- | --- | --- | --- | --- | --- |
| S | 1 vCPU · 2 GiB | $0.083/hr | $60/mo | 1 vCPU · 2 GiB | Developer $20/mo · $20 usage included |
| M | 2 vCPU · 4 GiB | $0.166/hr | $121/mo | 2 vCPU · 4 GiB | Developer $20/mo · $20 usage included |
| L | 4 vCPU · 8 GiB | $0.331/hr | $242/mo | 4 vCPU · 8 GiB | Developer $20/mo · $20 usage included |
| XL | 8 vCPU · 16 GiB | $0.662/hr | $484/mo | 8 vCPU · 16 GiB | Pro $250/mo · $250 usage included |

A month is 730 hours of running. Superserve bills CPU the whole time a sandbox runs, so it is always 100%.

Compute and memory only; storage is billed separately. Monthly plans include usage credit.

## Limits and features

Native provider capabilities, used directly:

| Feature | Value | Details |
| --- | --- | --- |
| Isolation | microVM | Firecracker |
| Snapshots | Memory + disk | and fork |
| Kept after stop | Disk + memory | on pause |
| When idle | Optional auto-pause |  |
| Max session | No limit | Developer / Pro |
| Max vCPU | 16 | Pro plan |
| Max memory | 64 GiB | Pro plan |
| Max disk | — |  |
| Sizing | Per template |  |
| Custom image | Yes |  |
| GPU | — |  |
| Self-hosting | BYOC | Enterprise, early partners |
| Max concurrency | 2,000 | Pro; Developer 100 |
| Secrets | Injected | on egress |
| Network controls | Yes | domains / CIDRs |
| Preview URLs | Yes | public / authenticated |

Pricing sources, checked 10 Nov 2026: [Superserve introduction and SDKs](https://docs.superserve.ai/introduction), [Pause, resume and lifecycle](https://docs.superserve.ai/sandbox/lifecycle), [Snapshots and forks](https://docs.superserve.ai/sandbox/snapshots), [BuildSpec and resource limits](https://docs.superserve.ai/templates/build-spec), [Outbound network rules](https://docs.superserve.ai/sandbox/networking), [Preview ports and authentication](https://docs.superserve.ai/sandbox/preview-urls), [Secrets and credential injection](https://docs.superserve.ai/secrets/overview), [Superserve plans and usage rates](https://www.superserve.ai/pricing/).

## Benchmarks

From [ComputeSDK](https://www.computesdk.com/benchmarks/sandboxes), which runs the same test against every provider. Latest run Nov 6, 2026.

- Burst TTI score: 93.2. Rank 10 of 30 on the latest run, down 1.9 from Oct 30. ([source](https://www.computesdk.com/benchmarks/sandboxes/burst-tti/superserve))
- Build time: 59.4 s. Rank 13 of 35 on the latest run, 4.9 s faster than Oct 30. Failed Oct 2. ([source](https://www.computesdk.com/benchmarks/sandboxes/dax/superserve))

| Run | Burst TTI score | Build time |
| --- | --- | --- |
| Nov 6, 2026 | 93.2 | 59.4 s |
| Oct 30, 2026 | 95.1 | 64.3 s |
| Oct 23, 2026 | 95.2 | 53.1 s |
| Oct 20, 2026 | 90.5 | 68.7 s |
| Oct 16, 2026 | 94.6 | 96.9 s |
| Oct 9, 2026 | 94.2 | 112.8 s |
| Oct 2, 2026 | 91.0 | Failed |
| Sep 25, 2026 | 91.0 | 94.1 s |