CreateOS Sandbox vs SF Compute
SF Compute fits ML teams that need 8-GPU H100 or H200 nodes for a defined window and want to avoid long fixed contracts. CreateOS Sandbox fits per-request code execution on CPUs, where waiting up to 10 minutes for a boot would not work. The two solve different problems.
Pricing at a Glance
CreateOS rates verified 17 October 2026. SF Compute pricing as published on 17 October 2026. Rates billed per GB rather than per GiB, or as bundled sizes, are quoted as published rather than converted.
CreateOS Sandbox
- Compute
- $0.0362 /vCPU-hr
- Memory
- $0.0116 /GiB-hr
SF Compute
Pricing not published: SF Compute's price list requires a login, and its docs show an example order of at most $20.00 per node-hour for an 8-GPU node, as of 17 October 2026.
Where CreateOS Sandbox Fits Better
What CreateOS Sandbox brings to this matchup, with a plain note on where SF Compute has the edge.
Per-second billing for runs of any length, where SF Compute sells node-hours.
Per-sandbox isolation, egress rules and pause, none of which SF Compute documents.
Public per-second rates you can check without an account.
Where SF Compute is strong
Lets teams buy GPU cluster time for exactly the hours they need at market prices and sell back unused reserved time, which suits bursty training runs.
Which to Pick
Pick CreateOS Sandbox if
- You need CPU sandboxes per request for agent code.
- You need egress rules and pause on each environment.
- You want to start on a free plan.
Pick SF Compute if
- You need 8x H100 or H200 nodes for a training run of known length.
- You want to resell reserved GPU time you do not use.
- You want to avoid long fixed GPU contracts.
Head-to-Head Comparison
Every dimension side by side, checked on 17 October 2026 against each vendor's own docs and pricing pages. "Not published" means we could not find it on the vendor's pages, not that the feature is missing.
| Dimension | CreateOS Sandbox | SF Compute |
|---|---|---|
| Built for | Running agent and untrusted code in isolated, disposable Linux sandboxes | Buying and reselling 8-GPU node time for training and inference |
| GPU | None; no GPU shapes | Yes: H100, H200 (A100 appears in SKU examples) |
| Isolation | Firecracker micro-VM with its own guest kernel | Full VM booted from an image; bare metal also offered |
| Billing | Per second while running: $0.03616363/vCPU-hr, $0.01159025/GiB-hr, storage $0.0001584/GB-hr | Market-priced per node-hour orders |
| Pause and resume | Yes: disk, memory and CPU state saved; vCPU stops billing, memory and storage still bill | Not published; instances end when the allocation ends |
| Outbound network | Open with no rules; deny-by-default once any host, IP, CIDR or wildcard rule exists, enforced on the host outside the VM | Not published |
| Create by API | Yes: REST API, CLI, MCP server and SDKs for TypeScript, Go, Python, Rust, C# and Java (all but TypeScript pre-1.0) | sf CLI; API not described |
Common Questions
How long does an SF Compute instance take to become usable?
SF Compute says allocation happens within about a minute of a start set to now, but image download and boot can take up to 10 minutes before SSH works (vendor-stated). Its instances also end when your purchased allocation ends.
Can I see SF Compute prices without an account?
Not the price list, which sits behind a login. The homepage shows illustrations such as $3.00 per GPU-hour reserved, and the docs use an order example of at most $20.00 per node-hour. CreateOS publishes per-second rates openly: $0.03616363 per vCPU-hour and $0.01159025 per GiB-hour.
What happens to running SF Compute instances when my allocation shrinks?
SF Compute terminates instances by priority, newest first, if your allocation drops below the number running. Purchased time is separate from instances, so you can recreate instances without losing time. CreateOS sandboxes run until you destroy or pause them.
Sources
Checked 17 October 2026.
See CreateOS Sandbox for yourself
Book a demo, or explore the Sandbox and its pricing.