CreateOS Sandbox vs Tama
If your agent needs a GPU, pick tama: it rents cards from T4 to B300 and restores a stopped machine with GPU memory and CUDA context intact. CreateOS Sandbox has no GPU, so it fits CPU agent and code-execution work where you want public per-second rates and forks that carry memory. For most workloads the GPU question settles the choice.
Pricing at a Glance
CreateOS rates verified 17 October 2026. tama 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
tama
Pricing not published on public pages; tama shows per-card hourly rates in its CLI after sign-in and bills per second while running, nothing while stopped, as of 17 October 2026.
Where CreateOS Sandbox Fits Better
What CreateOS Sandbox brings to this matchup, with a plain note on where tama has the edge.
CPU and memory rates are public: $0.03616363 per vCPU-hour and $0.01159025 per GiB-hour. tama shows its rates only in the CLI after sign-in.
Forks resume with memory, because a fork clones a paused sandbox's full snapshot. tama forks boot cold from a disk copy.
Documented outbound controls: open by default, then deny-by-default once any host, IP, CIDR or wildcard rule exists.
Where tama is strong
tama can stop a GPU machine mid-run and restore it warm in about a second (vendor-stated) with processes, GPU memory and CUDA context intact, from a wide NVIDIA catalog that includes B200, H200, H100 and RTX 5090, and it charges nothing while stopped.
Which to Pick
Pick CreateOS Sandbox if
- Your workload is CPU code execution for agents and you want to price it up front.
- You fork a prepared environment many times and need memory in each copy.
- You need egress allowlists enforced outside the VM.
Pick tama if
- Your agent or ML job needs a GPU.
- You want a stopped machine to cost nothing and come back with GPU state.
- You fine-tune or profile on a single card between agent runs.
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 | tama |
|---|---|---|
| Isolation | Firecracker micro-VM with its own guest kernel | Linux VM; hypervisor not named |
| Pause and resume | Saves disk, memory and CPU state; vCPU billing stops while paused, memory and storage still bill | Stop snapshots the machine and pauses billing; warm restore with processes and GPU memory if memory was captured |
| Fork | Fork a paused sandbox into many independent copies | Fork from a snapshot boots cold from a disk copy; memory not inherited |
| Outbound network default | 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 |
| Inbound access | Closed by default; opt-in HTTPS URL | tama expose gives an HTTPS URL open to anyone with the link; port-forward; SSH |
| Largest size | 8 vCPU / 16 GB on Enterprise; 8 / 8 on Pro, 4 / 4 on Beginner, 1 / 1 on Free | Not published (floor 1 vCPU / 1024 MiB; one GPU per machine) |
| Session length | No cap; optional idle auto-pause from 60 s to 24 h, off by default | No idle timeout or lifetime limit unless --ttl is set |
| Billing | Per second: $0.03616363 per vCPU-hour, $0.01159025 per GiB-hour, storage $0.0001584 per GB-hour; no egress fees (50 GiB bandwidth budget per sandbox) | Per second while running; nothing while stopped; rates only in CLI after sign-in |
| SDKs and interfaces | TypeScript, Go, Python, Rust, C#, Java (all but TypeScript pre-1.0), CLI, MCP, REST | CLI, Python SDK, TypeScript SDK, gRPC/Connect API |
| GPU | No GPU | Yes: B200, H200, H100, A100-80, L40S, RTX 5090, T4 and more; one card per machine |
Common Questions
Does CreateOS Sandbox offer GPUs like tama?
No. CreateOS has no GPU shapes; its largest sandbox is 8 vCPU / 16 GB on Enterprise. tama offers one NVIDIA card per machine, from T4 and L4 up to H200, B200 and B300, so GPU work belongs there.
What does a stopped tama machine cost compared with a paused CreateOS sandbox?
A stopped tama machine is billed nothing. A paused CreateOS sandbox stops vCPU billing but keeps billing memory and storage, because its full memory image is kept for resume. For machines that sit idle most of the day, tama's model is cheaper to hold.
Are tama's exposed URLs private?
No. tama expose publishes a port at an HTTPS URL that anyone with the link can open; for local-only access it offers port-forward over its authenticated channel. CreateOS keeps inbound access closed until you enable an HTTPS URL for the sandbox.
Sources
Checked 17 October 2026.
See CreateOS Sandbox for yourself
Book a demo, or explore the Sandbox and its pricing.