# Arker

Official pricing: https://arker.ai/docs/pricing  
Category: agent-sandbox · Isolation: vm

## Pricing regimes (raw)

- **AWS US VMs, on-demand run time (last public planning-table rates)** (sizes): small (2 vCPU / 8 GiB / 32 GiB) 2 vCPU/8 GiB $0.187736/h; large (4 vCPU / 16 GiB / 32 GiB) 4 vCPU/16 GiB $0.372016/h
- **AWS US pool, 12-month commitment (always-on, 19.55% off)** (sizes): small (2 vCPU / 8 GiB / 32 GiB) 2 vCPU/8 GiB $0.15104/h; large (4 vCPU / 16 GiB / 32 GiB) 4 vCPU/16 GiB $0.299299/h [sales]
- **AWS US pool, 24-month commitment (always-on, 22.83% off)** (sizes): small (2 vCPU / 8 GiB / 32 GiB) 2 vCPU/8 GiB $0.144871/h; large (4 vCPU / 16 GiB / 32 GiB) 4 vCPU/16 GiB $0.287075/h [sales]
- **AWS US pool, 36-month commitment (always-on, 48.56% off)** (sizes): small (2 vCPU / 8 GiB / 32 GiB) 2 vCPU/8 GiB $0.096581/h; large (4 vCPU / 16 GiB / 32 GiB) 4 vCPU/16 GiB $0.191384/h [sales]

## Features

Yes: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, whole disk survives a stop, volumes, 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, nested virtualization, root, systemd, browser, desktop GUI, computer-use API, code interpreter, browser automation API, desktop control API, browser + desktop control, preinstalled agents, SSH, HTTPS preview URLs, egress allowlist, open internet, private networking, BYOC, SSO, GPU, arm64, Windows, Python, Node.js, Go, Rust, GPU desktop, credential injection, HTTP method/path egress rules, wake on request, live resize, live fork (no pause), memory fork, EU data residency, automatic snapshots, snapshots on demand, fast boot (benchmarked), gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), secret proxy, secret proxy for any API, extra volumes, shared volumes, FUSE mounts, kernel features (eBPF, mounts, cgroups)

No: anti-bot stealth, CAPTCHA solving, residential IPs, public IPv4, static egress IP, self-hosting, open source, SOC 2, HIPAA, MCP server, usage per sandbox (API), agent harness API, hosted agent API (their own agent)

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

## Company

- Ownership: independent
- Stage: unknown [source](https://arker.ai/about)
- Accelerators: none found
- Total raised: unknown
- Last round: unknown (unknown) [source](https://arker.ai/about)
- Revenue: unknown

Very young company: GitHub org ArkerHQ created 2025-12-11; product/pricing docs only (rate card behind WorkOS login). About page names CEO Caleb Winston and 6 engineers (ex-Snowflake, AWS, NVIDIA, Stanford) and claims VC backing ('tier-1 venture capital firms and public company CEOs') but discloses no investor names, round size, or date. No TechCrunch/HN/Crunchbase coverage of funding was found; a YC company-page lookup for 'arker' returned 404. Buyers should treat funding/stage/revenue as undisclosed rather than absent.

## Caveats

- No public list rate card: arker.ai/pricing redirects to a WorkOS login, and on 2026-10-14 the docs 'Replicas capacity planning' table (1,000 always-on VMs x 730 h) that every on-demand/commit price here is derived from is no longer on arker.ai/docs/pricing.md. The prices are the last published ones; the only live public number is the pools.md example quote ($1,466.34 for 32 vCPU / 64 GiB / 256 GiB, aws us-west-2, 30 days), ~2.7% above them.
- Only two public shapes (small 2/8/32, large 4/16/32). A 4 vCPU / 8 GiB workload is priced on the large preset ($0.372/h) because nothing public prices 4/8.
- Derived (fragile) unit rates: solving small, large and the earlier pools-doc example ($1,428 for aws us-west, 32 vCPU / 64 GiB / 256 GiB, 30 days) gave ~$0.0301/vCPU-h, ~$0.0155/GiB-h RAM, ~$0.000108/GiB-h disk (~4/8 = $0.244/h). The live example is now amount_cents 146634 ($1,466.34, docs/api/pools.md, 2026-10-14). Not used in modes.
- 'Charges for run time, not machine time' is marketing text; whether idle/suspended VMs pay for memory/disk retention is not published (paused_ram_billed null). Prices exclude persistent storage; filesystem (object-storage volume) price unpublished.
- Commit modes are pools: private access, minimum 30-day term, prepaid on the next invoice, cannot be deleted; after expiry VMs keep running at normal rates. Modeled as always-on per-instance.
- EU Hetzner/Scaleway rows were explicitly "proposed prices, not currently published offers" and are removed from the pricing docs on 2026-10-14: flagged legacy so they are never priced.
- Egress, IPv4 and GPU prices are not published (GPU platforms x86_64-a100sxm-80gb and x86_64-h200sxm only in arker us-west per docs/regions.md, 2026-10-14; fractional vGPU from 0.125).
- No free tier. Billing granularity and minimums unknown.
- unverified: may bill run time only (active). Docs: "charge for run time, not machine time" (https://arker.ai/docs/pricing); the metering rule (what counts as run time, suspended-VM memory/disk) is unpublished, so modes keep alloc and the run_time_pct knob approximates it.
- Snapshot / persistent-storage price not published ("Prices exclude persistent storage and taxes", docs/pricing, 2026-09-28): snapshot_gib_month stays null, so retained state shows as unknown, not free.
- nested_virt is opt-in and placement-limited: fork API `nestedvirt` = "Give an `image` or `dockerfile` VM access to `/dev/kvm`. Unsupported placements return `400`." (arker.ai/docs/api/fork.md, checked 2026-09-28). Which placements support it and any price difference are not published.
- Windows exists (public 'windows' base VM, 1-2 vCPU / 8 GiB, arker eu and arker us-west only, docs/vms.md, 2026-10-14) but has no published price, and the priced modes are AWS US Linux presets, so a Windows workload stays unpriced.
- macOS (`m4` platform, Tart images) is documented in the fork API but not in the public regions table or VM catalog (2026-10-14): features.macos is null (unverified), likely private access.
- Regions changed (docs/regions.md, 2026-10-14): arker eu (graviton2, zen4), arker us-west (graniterapids, A100 80GB, H200), aws eu-north-1 (new, September 2026), aws us-east-1, aws us-west-2, gcp us-central1. Arker's own eu and us-west sites publish no city or country; arker eu is mapped to DE because the DPA lists Hetzner (EEA) and AWS (United States, EEA) as hosting subprocessors, and arker us-west has no state.

## How this provider charges


Arker (arker.ai) sells hyper-elastic VMs through a fork/run/sync API on its own hypervisor, placed on AWS, GCP or
Arker bare metal. **There is no public rate card.** arker.ai/pricing redirects to a login-gated console page. The only
public numbers are in the docs' "Replicas capacity planning" table: 1,000 always-on VMs × 730 h, in two shapes (small
2 vCPU / 8 GiB / 32 GiB and large 4 vCPU / 16 GiB / 32 GiB). Each has a 1-month (on-demand) price and 12/24/36-month
commitment prices. Arker says it charges for **run time, not machine time**: VMs scale to zero between runs and
policies can suspend them while waiting on LLM responses. The exact metering rule is not published.

Derived on-demand (AWS US): small **$0.1877/h**, large **$0.3720/h** (table ÷ 1,000 ÷ 730). No public 4/8 shape, so a
4 vCPU / 8 GiB workload is priced on "large".

## Regime table

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| On-demand, AWS US (1-month column) | Default self-serve account; no free tier | "Run time" of VMs; exact unit/granularity behind login | small $137,047 / large $271,572 per 1,000 VMs per month → $0.18774 / $0.37202 per VM-h | https://arker.ai/docs/pricing |
| Commit 12 months (pool) | Reserved capacity term; private access | Prepaid on next invoice, whole term, cannot be deleted | 19.55% off: $110,259 / $218,488 per 1,000/mo → $110.26 / $218.49 per VM-month | https://arker.ai/docs/pricing, https://arker.ai/docs/api/pools.md |
| Commit 24 months (pool) | same | same (AWS has no 2-y plan; "procurement ladder") | 22.83% off: $105,756 / $209,565 | same |
| Commit 36 months (pool) | same | same | 48.56% off: $70,504 / $139,710 | same |
| Commit formula | Any term 1–36 months | price × (1 − D(x)), x = (months − 1)/35 | AWS D = 1.754933379x³ − 2.505391256x² + 1.236007877x; Hetzner-EU D = 0.08593592x³ − 0.13996941x² + 0.13095657x; Scaleway-EU D = 0.040349144x³ − 0.079545455x² + 0.339196311x | https://arker.ai/docs/pricing |
| Pools (API) | Orgs with pools enabled (private access for partnered background-agent companies) | Quote → buy at `amount_cents`; min term 30 days; VMs placed via `pool_id`; after `ends_at`, VMs keep running at normal rates | Example quote: aws us-west, 32 vCPU / 64 GiB / 256 GiB, 30 days = $1,428 (illustrative) | https://arker.ai/docs/api/pools.md |
| Proposed EU on Hetzner | **Not offered**, proposed prices | Same table model | $22,074 / $51,556 per 1,000/mo (1 mo) → $0.0302 / $0.0706 per VM-h; 36 mo $20,376 / $47,591 | https://arker.ai/docs/pricing |
| Proposed EU on Scaleway | **Not offered** | Same | $22,603 / $45,205 (1 mo) → $0.0310 / $0.0619 per VM-h; 36 mo $15,822 / $31,644 | same |
| Suspend / scale-to-zero | VM idle between runs; `scaling: {suspend: true}` policy suspends while outbound response pending or between inbound requests | "Charge for run time, not machine time"; memory+disk persisted to NVMe then object storage | Retention/storage price unpublished | https://arker.ai/docs/pricing, https://arker.ai/docs/api/policies.md |
| Storage / filesystems | Persistent disk, object-storage filesystems | "Prices exclude persistent storage" | unpublished | https://arker.ai/docs/pricing |
| Private regions (VPC / on-prem), private networking | Contact sales | Control plane in cloud, customer compute | unpublished | https://arker.ai/docs/deployment.md |
| GPU (A100 40/80, H100, H200; fractional 0.125) | arker us-west, gcp us-central1 | unpublished | null | https://arker.ai/docs/regions.md |
| Free tier | — | None ("we don't currently offer a free tier") | $0 credit | https://arker.ai/docs/deployment.md |

Dated changes: Pools and AWS eu-north-1 region (Sep 2026); run/fork queueing (Aug 2026); egress and auto-scaling
policies (Jul 2026). The "public on-demand list prices as of 2026-06-11" comparison on the docs page covers competitors, not Arker.

## Gotchas

1. **No public list price.** Every Arker number here comes from a capacity-planning example. It may not match what the console quotes.
2. **Only two shapes are publicly priced**, and both are RAM-heavy (4 GiB/vCPU). A 4/8 workload pays for 16 GiB. Implied unit rates from the table plus the pools example (~$0.030/vCPU-h, ~$0.0155/GiB-h, ~$0.000108/GiB-h disk → 4/8 ≈ $0.244/h) are fragile back-solves.
3. **"Run time, not machine time"** is the headline differentiator, but the metering rule is not public: what counts as run time, and whether suspended VMs pay for memory/disk. Suspending during LLM waits could cut agent bills a lot. It could also be marketing.
4. **Commitments are prepaid pools that cannot be deleted.** They pay off only if VMs run close to 24/7. At 24% duty (8 h × 22 d), even the 36-month price is ~2x on-demand.
5. **The cheap EU rows are "proposed", not offered.** They are ~5x cheaper than AWS US.
6. The docs' competitor table puts boat.dev at $0.036/h for 4 shared vCPU / 8 GB. That is a competitor claim; boat.dev's own pages are the authority.

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent × 8 h/day × 22 days = **8,800 VM-hours**, 30% CPU, 50 GiB snapshots,
100 GiB egress. Priced on the "large" preset (4 vCPU / 16 GiB), the smallest public shape with 4 vCPU.

| Regime | Feasible? | Monthly total |
|---|---|---|
| On-demand AWS US, billed wall-clock | Yes (no published concurrency cap) | 8,800 × $0.372016 = **$3,273.74** + storage/egress (unpublished) |
| On-demand, VMs suspended 50% of session (run-time billing) | If the workload idles and suspend policies apply | $1,636.87 + unpublished retention cost |
| Derived unit rates (4/8 exactly, ~$0.244/h) | Speculative | ~$2,151 |
| 12-month pool, 50 large always-on | Private access | 50 × $218.488 = $10,924.40 |
| 36-month pool, 50 large always-on | Private access | 50 × $139.71 = $6,985.50 |
| Proposed EU Hetzner (not offered) | No | would be 8,800 × $0.070625 = $621.50 |
| Proposed EU Scaleway (not offered) | No | would be 8,800 × $0.061925 = $544.94 |

Snapshots (50 GiB) and egress (100 GiB): price not published, so excluded.