# Freestyle

Official pricing: https://www.freestyle.sh/pricing  
Category: agent-sandbox · Isolation: bare-metal-vm

## Pricing regimes (raw)

- **On-demand (allocated, per second)** (resource): $0.04032/vCPU-h, $0.0129/GiB-h
- **Enterprise active-CPU pricing** (resource) [sales]
- **Enterprise node-based (dedicated capacity, flat rate)** (resource) [sales]

## 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, full VM (own kernel), Docker inside, nested virtualization, root, systemd, preinstalled agents, SSH, HTTPS preview URLs, custom domains, raw TCP inbound, egress allowlist, open internet, static egress IP, private networking, SSO, Python, Node.js, credential injection, HTTP method/path egress rules, wake on request, live resize, live fork (no pause), memory fork, audit logs, scoped API keys, 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

No: volumes, 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, self-hosting, BYOC, open source, GPU, arm64, Windows, macOS, GPU desktop, webhooks, EU data residency, MCP server, agent harness API, hosted agent API (their own agent), extra volumes, shared volumes

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

## Caveats

- free.monthly_credit 18.38 = dollar value of per-resource allowances (200 vCPU-h $8.06 + 400 GiB-h RAM $5.16 + 60,000 GiB-h storage $5.16). The allowances are NOT fungible: each only offsets its own resource. They apply on every plan before the fee credit, and on Free they are hard caps.
- egress_free_gib 500 applies to Hobby/Pro; Free includes 50 GB and stops there. Transfer counts bytes in BOTH directions (ingress too). VPC-internal traffic is free.
- disk_gib_month = $0.000086/GiB-h x 730. Disk is billed on allocated size (default 32 GB) while running, paused or stopped. Ephemeral VMs (autoDeleteSeconds 0) stop billing disk when deleted. There is no per-sandbox free disk, but the 60,000 GiB-h monthly allowance covers ~82 GiB kept all month.
- Retained state priced at the storage rate ($0.000086/GiB-h = $0.06278/GiB-month): docs say "A paused VM holds no compute reservation, so it bills only for storage while it sleeps." Named-snapshot storage has no separate published price (only count limits 10/1,000/12,000); treated as the same storage rate (unverified). Whether a paused VM's saved RAM image counts as storage is undocumented.
- No default idle timeout: VMs bill until paused/stopped/deleted. idleTimeoutSeconds is network-activity based, so CPU-only jobs can be paused. Marketing's ~30 s auto-suspend is not in the docs.
- Enterprise active-CPU, node-based and commitment-discount rates are unpublished (modes left null).
- Allowances were DAILY on an older pricing page (20 vCPU-h/40 GiB-h/16,800 GiB-h storage per day); they are now monthly (200/400/60,000). The change date is unknown.
- No dedicated IPv4: every VM gets a public IPv6, and all VMs in an account share one outbound IPv4.
- Regions not documented; 'us' inferred from the Latitude.sh San Jose peering note.
- cold_start_ms: no computesdk cold-start run for freestyle (vendor claims ~65 ms); dax never succeeded in benchmarks.
- free.monthly_credit ($18.38) values separate monthly resource allowances (200 CPU-h, 400 GiB-h RAM, 60,000 GiB-h disk per the 2026-09-24 capture); it is not a pooled spendable credit, so unused disk allowance cannot offset CPU in reality. A 2026-07-08 capture advertised per-day allowances (20 CPU-h / 40 GiB-h / 16,800 GiB-h disk); do not reuse current allowances for historical estimates. https://web.archive.org/web/20260708095940/https://www.freestyle.sh/pricing , https://web.archive.org/web/20260924112048/https://www.freestyle.sh/pricing
- Order of credits is now documented: the per-resource allowances are applied first, and the Hobby/Pro fee is a floor on the remaining usage. Bill = max(fee, usage above allowances). An engine that computes max(fee, usage) - free.monthly_credit understates Hobby/Pro bills when usage is below the fee (e.g. Hobby at low usage: real $50, engine up to $31.62).
- Base snapshot presets (boot sizes): busybox 1/0.125 GiB/1 GB, ubuntu-sm 2/4/16, ubuntu 4/8/32 (default), ubuntu-lg 8/16/64, ubuntu-xl 16/32/128, ubuntu-2xl 32/64/128, ubuntu-3xl 64/128/256 (Enterprise only). Free tops out at ubuntu, Hobby at ubuntu-lg, Pro at ubuntu-2xl. Any valid size up to the plan max is allowed, so resource pricing (vcpu_options null) stays correct.

## How this provider charges


As of 2026-09-28. Product: Freestyle VMs (full KVM VMs on bare metal, memory snapshots, pause/resume).
Primary sources: https://www.freestyle.sh/pricing, https://www.freestyle.sh/docs/vms/pricing-and-limits,
https://www.freestyle.sh/docs/vms/lifecycle, https://www.freestyle.sh/products/vms,
https://apis.io/plans/freestyle-sh/freestyle-sh-plans-pricing/ (mirror of an older pricing page).

List rates (identical on every plan): vCPU $0.04032/h, memory $0.0129/GiB-h, storage $0.000086/GiB-h
(= $0.0628/GiB-month), data transfer $0.02/GB. Metered per second, billed per resource-hour, on ALLOCATION.
Default 4 vCPU / 8 GiB / 32 GB = $0.26448/h compute (+$0.00275/h disk).

## Regimes

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| Monthly allowance (all plans) | Every plan incl. Free, applied before anything else | First N resource-hours each month free; per-resource, not fungible | 200 vCPU-h ($8.06), 400 GiB-h RAM ($5.16), 60,000 GiB-h storage ($5.16), transfer 50 GB Free / 500 GB Hobby+Pro | /pricing, /docs/vms/pricing-and-limits |
| Free - hard cap | Free plan | Allowances are hard stops; exhausting compute pauses VMs until monthly reset, no auto-resume, no overage possible | $0; 10 concurrent, 10 saved VMs, 10 snapshots; per-VM max 4 vCPU/8 GiB/32 GB; account 40 vCPU/80 GiB/320 GB; no custom sizing. 200 vCPU-h = only 50 h of a default VM | /pricing |
| Hobby - min spend as credit | Hobby | Pay max($50, usage above allowances) | $50/mo fully usable as usage; 40 concurrent, 200 saved, 1,000 snapshots; 8 vCPU/16 GiB/64 GB per VM; 800 vCPU/1.56 TiB/3,125 GB totals | docs pricing-and-limits |
| Pro - min spend as credit | Pro | Pay max($500, usage above allowances) | $500/mo; 400 concurrent, 4,000 saved, 12,000 snapshots; 32 vCPU/64 GiB/256 GB per VM; 16,000 vCPU/31.25 TiB/62,500 GB totals | docs pricing-and-limits |
| Running (on-demand) | VM running, starting or pausing | vCPU + RAM + disk on allocated size, per second; no discount for idle-but-running | $0.04032/vCPU-h, $0.0129/GiB-h, $0.000086/GiB-h | docs pricing-and-limits |
| Always-on (no discount) | idleTimeoutSeconds null (the default - no idle timeout unless you set one) | Same as running, 730 h/month | 4/8 VM: $193.07/mo compute + $2.01 disk (32 GB) | derived |
| Paused | vm.pause() or idle timeout (network-inactivity based) | No compute; storage only. Doesn't count toward concurrency, does count toward saved VMs + disk budget | Disk $0.000086/GiB-h; saved-memory image billing not stated | docs lifecycle, pricing-and-limits |
| Stopped (persistent) | Persistent VM powered off | Disk only, memory discarded | $0.000086/GiB-h | docs lifecycle |
| Ephemeral | autoDeleteSeconds: 0 | VM deleted when it stops - no disk after the session | running rates only | docs pricing-and-limits |
| Snapshots | Named snapshots; kept until deleted | Storage price NOT published (count-limited: 10/1,000/12,000). Pricing page says persistent snapshots are paid-only | unknown (assumed storage rate in the example) | /pricing |
| Data transfer | Bytes crossing the DC boundary in either direction | $0.02/GB after 50 GB (Free, hard stop) / 500 GB (paid). VPC-internal free | $0.02/GB | docs pricing-and-limits |
| IPv4 | - | Not offered: per-VM public IPv6; all VMs share one outbound IPv4 | n/a | features/freestyle.json |
| Enterprise active CPU | Enterprise | Bill measured CPU time instead of allocated vCPU | rate unpublished | docs #enterprise-options |
| Enterprise node-based | Enterprise | Flat rate for dedicated capacity instead of per-resource billing | rate unpublished | docs #enterprise-options |
| Enterprise commitment | Enterprise | Reduced rates for usage/spend commitment | discount unpublished | docs #enterprise-options |
| Historical: daily allowance | Superseded, date unknown | Same rates; allowances were 20 vCPU-h, 40 GiB-h, 16,800 GiB-h storage PER DAY (~600/1,200/504,000 per month), plus repo/run quotas from the older Git/serverless product | - | apis.io mirror |
| Card verification | Adding card to Free | $1 charged then refunded; not credit | $1 refunded | docs pricing-and-limits |

## Gotchas

- Allowances are applied BEFORE the plan-fee credit on paid plans, so a Hobby account gets ~$18.38 of resource-hours + 500 GB transfer
  on top of the $50 credit. But they are per-resource: unused storage-hours cannot pay for vCPU-hours.
- The compute allowance got smaller: the older page gave 20 vCPU-h per DAY; now it is 200 per MONTH (~3x cut, date unknown).
- No self-serve always-on / reserved discount. The only discounts (active-CPU, node-based, commitment) are Enterprise with unpublished rates.
- Billing is on allocation: "a reserved core is yours whether or not the guest is busy". At 30% CPU utilisation you still pay 100%.
- There is no default idle timeout: a VM runs, and bills, until you pause it, set idleTimeoutSeconds/maxRunSeconds, or delete it.
  Marketing implies ~30 s auto-suspend; the docs don't. The idle timeout is based on network activity, so CPU-only work can get
  paused mid-job, while a VM that only prints to a PTY can still idle out.
- Marketing says "pay nothing while paused"; the docs say a paused VM "bills only for storage". Whether the saved RAM image counts as
  storage is not documented.
- Resize is up-only (live hot-add). Once grown, a VM can't shrink. VMs always start at their snapshot's size, so a snapshot taken
  from a big VM forces every clone to be big.
- Disk budget, not the saved-VM count, is the binding limit: Hobby fits only 97 default (32 GB) VMs despite "200 saved VMs".
- Concurrency counts running/starting/pausing VMs only, so paused VMs are "free" against the concurrency cap. 50 concurrent forces
  Pro ($500 minimum), because Hobby is capped at 40.
- Transfer counts both directions. Ingress (npm/pip pulls, uploads) uses up the 500 GB and then bills at $0.02/GB.
- Free contradicts itself: the pricing page shows persistent VMs/snapshots as unavailable on Free, but the docs limits table shows
  persistent VMs on every plan.
- Downgrading pauses VMs that are too large (or persistent) for the new plan, and they don't auto-resume. On Free, running out of
  allowance pauses VMs the same way.
- Snapshot storage has no published price, only count limits.
- exec() times out at 5 min. Longer jobs need PTY/nohup, which affects how you'd script auto-pause.

## Worked example

4 vCPU / 8 GiB (the default size, fits every plan's per-VM cap). 50 concurrent × 8 h/day × 22 days = 8,800 VM-h =
35,200 vCPU-h + 70,400 GiB-h RAM. CPU utilisation is 30%, but that is irrelevant here because billing is on allocation.
50 GiB of snapshots retained all month. 100 GiB egress. Month = 730 h. Disk = default 32 GB per VM.
Plan: 50 concurrent > Hobby's 40, so **Pro** ($500 minimum, credited against usage). Free: not eligible (10 concurrent, hard caps).

Common pieces:
- vCPU: (35,200 − 200) × 0.04032 = $1,411.20
- RAM: (70,400 − 400) × 0.0129 = $903.00
- Compute subtotal: **$2,314.20** (without the allowance it would be $2,327.42)
- Snapshots: 50 × 730 = 36,500 GiB-h. The price is unpublished; here it's assumed billed like storage at $0.000086. If snapshots are free, subtract ≤ $3.14.
- Egress: 100 GB < 500 GB included, so $0. Ingress counts too.

| Regime | Storage GiB-h (after 60,000 free) | Storage $ | Total/month |
|---|---|---|---|
| A. Pro, ephemeral VMs per session (autoDeleteSeconds 0), clone from snapshot | 50×32×176 = 281,600 + 36,500 − 60,000 = 258,100 | $22.20 | **$2,336.40** (usage > $500 min) |
| B. Pro, persistent VMs paused between sessions (disk kept 730 h) | 50×32×730 = 1,168,000 + 36,500 − 60,000 = 1,144,500 | $98.43 | **$2,412.63** (+ unknown saved-RAM storage, if billed) |
| C. Pro, VMs never paused (no idle timeout set = always-on) | same as B | $98.43 | compute (146,000−200)×0.04032 + (292,000−400)×0.0129 = $9,640.30. **$9,738.73** |
| D. Hobby | not eligible (40-concurrent cap). Rates are identical, so Hobby only differs when usage < $500 | - | n/a |
| E. Free | not eligible (10 concurrent; 200 vCPU-h hard stop) | - | n/a |
| F. Enterprise active CPU | rate unpublished | - | **null**. Illustrative only: if list rates applied to 30% of vCPU-h, (10,560−200)×0.04032 + $903.00 + $22.20 = $1,342.92. This is NOT a published price |
| G. Enterprise node-based / commitment | rates unpublished | - | **null** |
| H. Historical daily-allowance regime (superseded) | daily storage allowance 16,800 GiB-h > 14,000 GiB-h/day used, so free | $0 | (1,600−20)×22×0.04032 + (3,200−40)×22×0.0129 = **$2,298.33** |

The big swing is lifecycle (A vs C = 4.2x), not plan choice. Always set idleTimeoutSeconds/maxRunSeconds or pause explicitly.