# tama

Official pricing: https://tama.computer/  
Category: agent-sandbox · Isolation: vm

## Pricing regimes (raw)

- **CPU on-demand** (resource): $0.01166/vCPU-h, $0.005523/GiB-h
- **RTX 4090 live bundle** (sizes): RTX 4090 6 vCPU/17 GiB $0.54/h
- **A 6000 live bundle** (sizes): A 6000 28 vCPU/58 GiB $0.675/h
- **RTX 5090 live bundle** (sizes): RTX 5090 7 vCPU/64 GiB $0.8775/h
- **A 4000 live bundle** (sizes): A 4000 8 vCPU/45 GiB $1.08/h
- **L 40 live bundle** (sizes): L 40 26 vCPU/192 GiB $1.188/h
- **L 40 S live bundle** (sizes): L 40 S 12 vCPU/72 GiB $1.188/h
- **L 4 live bundle** (sizes): L 4 8 vCPU/48 GiB $1.2825/h

## Features

Yes: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, idle auto-stop, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, full VM (own kernel), Docker inside, desktop GUI, preinstalled agents, SSH, HTTPS preview URLs, GPU, Python, Node.js, live fork (no pause), automatic snapshots, snapshots on demand, agent harness API, gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables)

No: volumes, browser, computer-use API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, public IPv4, egress allowlist, static egress IP, self-hosting, BYOC, open source, SOC 2, HIPAA, SSO, arm64, Windows, macOS, credential injection, HTTP method/path egress rules, wake on request, live resize, memory fork, EU data residency, MCP server, hosted agent API (their own agent), secret proxy, secret proxy for any API, extra volumes, shared volumes

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/tama.json → feature_evidence

## Caveats

- Storage, snapshot GiB-month and public IPv 4 prices are not published in the reviewed sources; null is not free.
- GPU prices are live from-offers and changed between search cache and render: use renderedA 6000$.68/28 CPU/58 GB andA 100-40 GB $2.69/30 CPU/200 GB, not cached $.77/$1.89.
- Whole-node rates include multiple cards; do not multiplyH 200$43.20 by 8 again. Docs initially say checkpointing only one card, while offers list multi-card minima:capability/launch conflict requires confirmation.
- Default idle timeout is disabled;--ttl minimum 60 seconds is opt-in. Files only persist under /workspace. Warm resumes retain RAM; fork descriptions guarantee baseline disk, not all process memory.
- Homepage calculator says $20 ofcredit buys 220 hours; that is not proof of a free $20 signup grant. Network pricing, concurrency andsnapshot retention unverified.
- Desktop: llms.txt documents an optional virtual desktop (`tama new --desktop`, "these machines resume cold"; SDK tama.desktop() returns a browser desktop URL) and a browser terminal. No managed browser/CDP endpoint or computer-use (screenshot/click) API is documented, so browser and computer_use are false.
- Egress and extra-disk prices: still not published (llms.txt / docs / offers API re-checked 2026-09-28; /workspace is a fixed durable 50 GiB disk, no resize price). Left null.
- Live offers drift: on 2026-09-28 the offers API briefly showed H100 $3.375/h and L40S $2.295/h; on 2026-09-29 it is back to H100 $3.6855 and L40S $1.188 (= card). H200/B200/Gaudi 2 are still absent from the live API and live rate card (only in the site's static fallback catalogue). A16 (2-card box $1.377/h) and V100/V100-32 ($3.159/h) are live offers, not discontinued.
- GPU machines are one card each: '--gpu-count' >1 fails at create because checkpointing more than one card is unsupported. Multi-card nodes (A16 2-card box $1.377/h) are billed whole. Whole-node rentals are 'billed whole for as long as your machine holds it' and may need 'minutes' to provision; serverless-pool cards are per card-hour on shared capacity. --capacity auto|serverless|on-demand exists but the offers API returns one price per card.
- Offers API 2026-09-29: RTX4090 and RTX5090 rows have available_nodes 0 and node size 0 (probably serverless-pool cards with no warm workers, not sold out); an A10G row exists with hourly_micros 0 and cards_on_node 0 (placeholder, not orderable). llms.txt also names B200, H200, T4, GH200, B300 as cards, but 'tama offers is the list that is true today' and none of these are in it.
- StarSling's 2026-09-29 benchmark run detected an OCI container (not a VM) inside this provider's 4 vCPU / 8 GB sandbox; the card keeps the provider's documented isolation (vm) until confirmed.

## How this provider charges


As of 2026-09-28. USD unless explicitly stated; public list prices unless a row is marked promotion, historical, or quote-only.

## Regime table

| Regime | When it applies | How billed | Numbers / terms | Source |
|---|---|---|---|---|
| CPU | Running allocation | Per second | $.0117/vCPU-h+$.0055/GiB-h;1–32 CPU,1–128 GB | https://tama.computer/ |
| GPU | Live provider offers | Per second | Cards with serverless pools price per card; other offers whole node; see gpu array and saved live grid | https://tama.computer/ |
| Stopped/snapshot | Pause or snapshot | No charge | 50 GiB durable/workspace; warm memory or cold disk restoration | https://tama.computer/llms.txt |
| Whole GPU node | Multiple-card minimum | Whole-node price | A 16 minimum 2,$1.38/h; H 200 minimum 8,$43.20/h in rendered offer grid | https://tama.computer/ |
| Pay per running second | Account plan | Monthly fee / commitment | $0; concurrency unpublished/custom; No published subscription fee or concurrency guarantee; current offers vary. | https://tama.computer/ |

## Gotchas

- GPU prices are live from-offers and changed between search cache and render: use renderedA 6000$.68/28 CPU/58 GB andA 100-40 GB $2.69/30 CPU/200 GB, not cached $.77/$1.89.

- Whole-node rates include multiple cards; do not multiplyH 200$43.20 by 8 again. Docs initially say checkpointing only one card, while offers list multi-card minima:capability/launch conflict requires confirmation.

- Default idle timeout is disabled;--ttl minimum 60 seconds is opt-in. Files only persist under /workspace. Warm resumes retain RAM; fork descriptions guarantee baseline disk, not all process memory.

- Homepage calculator says $20 ofcredit buys 220 hours; that is not proof of a free $20 signup grant. Network pricing, concurrency andsnapshot retention unverified.

- Storage, snapshot GiB-month and public IPv 4 prices are not published in the reviewed sources; null is not free.

## Browser and desktop dimensions

```json
{
  "hour_usd": null,
  "included_hours": null,
  "proxy_residential_gb_usd": null,
  "proxy_datacenter_gb_usd": null,
  "captcha_solve_usd": null,
  "stealth_surcharge_usd": null,
  "recording_storage_gib_month_usd": null,
  "recording_retention": null,
  "persistent_profiles": null,
  "resolution": null,
  "gpu_desktops": "GPU compute public; desktopbrowser stack not guaranteed",
  "os_options": [
    "linux"
  ],
  "browsers": null
}
```

## Worked example

Requested: 4 vCPU / 8 GiB; 50 concurrent × 8 hours/day × 22 days = **8,800 runtime hours**, 1,100 eight-hour starts if recreated daily; 30% CPU; 50 GiB-month snapshots; 100 GiB egress.

CPU-only 4/8 rate $.0908/h;8,800 h costs **$799.04** before network.30%CPU does not discount allocated rates. Built-in snapshots/stopped machines are advertised free, but 50 GiBsnapshot retention and 50 concurrency entitlement need confirmation. GPU offers are separate bundles; no need for a GPU in this example.

## Sources

- https://tama.computer/

- https://tama.computer/llms.txt

- https://tama.computer/docs