# Tenki Sandbox

Official pricing: https://tenki.cloud/pricing  
Category: agent-sandbox · Isolation: vm

## Pricing regimes (raw)

- **Sandbox session (allocated vCPU + RAM per second while running)** (resource): $0.0504/vCPU-h, $0.0162/GiB-h
- **Tenki Runners, Linux x64 GitHub Actions (per core-minute)** (sizes): tenki-standard-small-2c-4g 2 vCPU/4 GiB $0.24/h; tenki-standard-medium-4c-8g 4 vCPU/8 GiB $0.48/h; tenki-standard-large-8c-16g 8 vCPU/16 GiB $0.96/h; tenki-standard-large-plus-16c-32g 16 vCPU/32 GiB $1.92/h
- **Tenki Runners, macOS (Apple M4 Pro) GitHub Actions (per core-minute)** (sizes): tenki-macos-26-mini 2 vCPU/4 GiB $2.4/h; tenki-macos-26-small 4 vCPU/8 GiB $4.8/h; tenki-macos-26-medium 6 vCPU/16 GiB $7.2/h; tenki-macos-26-large 8 vCPU/32 GiB $9.6/h

## Features

Yes: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, volumes, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, full VM (own kernel), Docker inside, root, systemd, preinstalled agents, SSH, HTTPS preview URLs, egress allowlist, open internet, static egress IP, SOC 2, arm64, macOS, Python, Node.js, Go, memory fork, spend limits, MCP server, automatic snapshots, snapshots on demand, fast boot (benchmarked), gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), extra volumes

No: idle auto-stop, your own Docker/OCI image, browser, desktop GUI, computer-use API, code interpreter, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, public IPv4, raw TCP inbound, self-hosting, BYOC, open source, HIPAA, SSO, GPU, Windows, GPU desktop, HTTP method/path egress rules, wake on request, live resize, live fork (no pause), EU data residency, agent harness API, hosted agent API (their own agent), secret proxy, secret proxy for any API, shared volumes

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

## Caveats

- One credit pool per workspace is shared by sandboxes, CI runner minutes and code reviews ($1.00/review); plan concurrency (5 / 100) is also shared between runner jobs and sandbox sessions.
- Two storage meters: 'Sandbox Disk' $0.00000003/GiB-s (= $0.000108/GiB-h, $0.07884/GiB-month at 730 h) beyond 5 GiB free per session, presumably while the session exists/runs (engine charges per month); and 'Storage' $0.20/GiB-month from the first byte ($0.000000076/GiB-s on the pricing page) for snapshots, volumes, runner caches and paused-session state, one pool per workspace capped at 100 GiB (Starter) / 1,000 GiB (Team).
- Whether 'Sandbox Disk' keeps billing while a session is paused is undocumented; card assumes paused state is billed only via the $0.20 storage pool (disk_billed_when_stopped false).
- No idle auto-pause: the --idle-timeout option was removed on 2026-09-18 because it 'never had any effect'. Non-sticky sessions pause at their max duration (since 2026-09-11); the default max duration value is unpublished. Sticky sessions bill until terminated.
- Pricing-page plan cards now show '$100 sign-up credit' for Starter, but the pricing-page FAQ still says 'Starter's $10/month goes to your oldest workspace'; docs (llms-full billing section) say the $100 sign-up credit is a one-time grant per account (first workspace, no card) and pre-existing workspaces 'no longer receive the $10 monthly Starter credit'. Card uses $100 one-time.
- Team is $250/month billed monthly or $200/month billed yearly (docs pricing table); the pricing page shows $200 with '20% Saved'. Team's $100/month credit is separate from the fee (fee_is_credit false) and does not roll over.
- Other one-time credits: $10 for sharing a Tenki Runners post (one per user); startup program up to $50K credits and grants.
- Runner modes are CI minutes (product class ci-runner, not flagged alt): they rank in CI questions and show as a different kind of product elsewhere. Runner RAM and disk are bundled in the per-core price, so $/h is higher than the sandbox rate for the same 4c/8g ($0.48 vs $0.3312).
- Egress/ingress pricing not published (null): re-checked 2026-09-28 on https://tenki.cloud/pricing and https://tenki.cloud/docs/pricing.md, neither lists any network/egress charge (not stated as free either). No public IPv4; HTTPS preview URLs (20 per workspace plan) and SSH.
- Region: docs mention a default 'US region' for the SSH gateway; other regions not documented. Owned hardware (Luxor Technology); hypervisor not named.
- Starter per-session limit 4 cores / 8 GB; Team 16 / 64 GB; runners on Team only up to 16c/32g.
- Outbound allowlists (features.egress_allowlist) are not self-serve: 'Tenki must enable session egress policies for your workspace first' (contact support).
- Named preview-URL slugs (a stable wildcard under *.sb.tenki.sh) are Team-plan only, one per workspace.

## How this provider charges


Tenki (owned by Luxor Technology) sells three products from one per-workspace credit pool:

- **Tenki Sandbox**: full Linux VMs, billed per second at E2B-identical allocation rates.
- **Tenki Runners**: GitHub Actions CI minutes on x64 and macOS, a different product (flagged `alt`).
- **Code Reviewer**: $1.00 per review.

Plans set limits and included credits, and the concurrency limit is shared between runner jobs and sandbox sessions.

Base sandbox rates: vCPU $0.000014/s = **$0.0504/h**; RAM $0.0000045/GiB-s = **$0.0162/GiB-h**; sandbox disk
$0.00000003/GiB-s ($0.000108/GiB-h) beyond **5 GiB free**; storage pool **$0.20/GiB-month** from the first byte
($0.000000076/GiB-s) for snapshots, volumes, runner caches and paused-session state. 4 vCPU / 8 GiB = **$0.3312/h**.

## Regime table

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| Starter | Default for every workspace | $0 fee. **$100 one-time sign-up credit** (first workspace only), then PAYG from prepaid balance (top-up min $20, purchased credits never expire) | 5 concurrent, 5 members, 100 GiB storage, sandbox max 4 cores / 8 GB | https://tenki.cloud/docs/pricing.md, https://tenki.cloud/docs/account/billing |
| Team, monthly | Need >5 concurrent / bigger VMs | $250/mo **fee + separate** $100/mo credits (no rollover). Upgrading **drains** remaining Starter credit | 100 concurrent, 50 members, 1,000 GiB storage, 16 cores / 64 GB | docs pricing table; pricing FAQ |
| Team, annual | Same, billed yearly (12-month commitment; card flags it `sales` so it is opt-in in the estimator) | $200/mo ($2,400/yr), "20% saved" | as Team | https://tenki.cloud/pricing, https://tenki.cloud/docs/pricing.md |
| Enterprise | Beyond Team | Discounted rates, larger machines, unlimited concurrency on request, on-demand dedicated servers (custom), SLA | unpublished | docs pricing; billing docs |
| Sandbox running | Session `RUNNING` (also CREATING→RUNNING after a timed-out wait) | Allocated vCPU + RAM per second; "no minimums or per-seat fees" | $0.0504 / $0.0162 | pricing page |
| Sandbox disk | Root disk size (default 5 GiB; 20 GiB recommended for coding agents) | Per GiB-second beyond 5 GiB free | $0.000108/GiB-h | pricing page; sessions docs |
| Paused | `pause`, or reaching max duration (since 2026-09-11) | No compute; not counted toward concurrency; preserved memory+disk counts against the **storage pool** | $0.20/GiB-mo | concepts docs; changelog |
| Sticky session | `sticky: true` | Runs, and bills, until terminated. Ignores max duration | allocation rate | sessions docs |
| Idle | — | **No idle auto-pause**: `--idle-timeout` was removed 2026-09-18 because it "never had any effect" | — | changelog 2026-09-18 |
| Snapshots / volumes | Named memory+disk snapshots; volumes | Workspace storage pool | $0.20/GiB-mo; pool cap 100 / 1,000 GiB | docs pricing |
| x64 runners (alt) | GitHub Actions `tenki-standard-*` | Per second of job runtime (started second rounded up); queue free; RAM+disk bundled | $0.002/core-min = $0.12/core-h: 2c $0.24/h, 4c $0.48/h, 8c $0.96/h, 16c $1.92/h | https://tenki.cloud/docs/runners/sizes.md |
| macOS runners (alt) | `tenki-macos-*` (M4 Pro) | Same | $0.020/core-min = $1.20/core-h: 2c $2.40/h … 8c $9.60/h | pricing page |
| Runner cache | Beyond included | Same storage pool | $0.20/GB-mo | pricing page |
| Code reviews | Per review | — | $1.00 | pricing page |
| Other credits | Startup program; social-share reward | One-time | up to $50K; $10 (one per user) | pricing page; docs |
| Egress / IPv4 | — | Not published; no public IPv4 (HTTPS preview URLs, SSH) | null | — |

Dated changes: new pricing model (PAYG + Team) 2026-05-01. Sign-up credit replaced the legacy $10/month Starter credit
(date not stated). Sessions pause at max duration from 2026-09-11. Idle-timeout option removed 2026-09-18.

## Gotchas

1. **Pricing-page FAQ is stale.** It says "Starter's $10/month goes to your oldest workspace". The docs say the $10 monthly credit is discontinued and replaced by a $100 one-time sign-up credit.
2. **Team fee is not credit.** $250 buys limits plus a *separate* $100 credit, so Team costs $150/mo net of credit at zero usage. Upgrading forfeits leftover Starter credit.
3. **Concurrency is shared with CI.** 100 concurrent on Team covers runner jobs *and* sandboxes. A busy CI day can starve agents (jobs queue, they don't fail).
4. **The storage pool is capped.** Paused sessions keep memory+disk in the pool. 50 paused 8 GiB VMs with 20 GiB disks (~1,400 GiB) exceed Team's 1,000 GiB cap.
5. **No idle detection.** A non-sticky session runs until its max duration (default unpublished), then pauses. A sticky one bills until you terminate it. A create call that times out can still reach RUNNING and bill.
6. **Two disk meters**: per-session "Sandbox Disk" (5 GiB free) plus the pooled $0.20/GiB-month "Storage". Whether sandbox disk keeps billing while paused is undocumented.
7. **Runners cost more per hour than sandboxes** for the same shape (4c/8g: $0.48 vs $0.3312), because RAM and disk are bundled per core.

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent × 8 h/day × 22 days = **8,800 sandbox-hours**, 30% CPU, 50 GiB snapshots,
100 GiB egress; paused between shifts.

Compute: 8,800 × 0.3312 = **$2,914.56** (30% CPU changes nothing). Snapshots: 50 × $0.20 = **$10**. Default 5 GiB
root disk is free (a 20 GiB disk adds 15 × 0.000108 × 8,800 = $14.26). Paused-state storage (≈ 50 × 13 GiB = 650 GiB for
~554 paused h): ≈ **$98.5**. Egress: unpublished.

| Regime | Feasible? | Monthly total |
|---|---|---|
| Starter | No: 5 concurrent (sessions above the limit queue) | n/a |
| Team monthly | Yes | **$3,074.56** = 250 − 100 + 2,914.56 + 10 (≈ $3,173 with paused state in the pool) |
| Team annual | Yes | **$3,024.56** (≈ $3,123 with paused state); $2,400/yr prepaid |
| Enterprise | Yes | discounted, unpublished |
| Runners (alt): same hours on `tenki-standard-medium-4c-8g` | Only for CI jobs, not interactive | 8,800 × 0.48 = $4,224 + Team fee |
| Anti-pattern: sticky sessions left running 24 h/day | Yes | 150 + 50 × 24 × 22 × 0.3312 = **$8,893.68** |