# Daytona

Official pricing: https://www.daytona.io/pricing  
Category: agent-sandbox · Isolation: container

## Pricing regimes (raw)

- **Sandbox, container class (default)** (resource): $0.0504/vCPU-h, $0.0162/GiB-h
- **Linux VM sandbox (same rates; KVM / nested virt per-org feature)** (resource): $0.0504/vCPU-h, $0.0162/GiB-h
- **GPU sandbox, on-demand** (resource): $0.0504/vCPU-h, $0.0162/GiB-h
- **GPU sandbox, spot / preemptible** (resource): $0.0504/vCPU-h, $0.0162/GiB-h [spot]

## Features

Yes: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, volumes, idle auto-stop, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, Docker inside, nested virtualization, root, browser, desktop GUI, computer-use API, code interpreter, desktop control API, browser + desktop control, preinstalled agents, SSH, HTTPS preview URLs, custom domains, egress allowlist, open internet, private networking, BYOC, SOC 2, HIPAA, SSO, GPU, Windows, Python, Node.js, credential injection, live resize, memory fork, webhooks, audit logs, EU data residency, scoped API keys, MCP server, automatic snapshots, snapshots on demand, fast boot (benchmarked), inbound access rules, firewall inside (nftables), secret proxy, secret proxy for any API, extra volumes, shared volumes

No: full VM (own kernel), anti-bot stealth, CAPTCHA solving, residential IPs, public IPv4, raw TCP inbound, static egress IP, self-hosting, open source, arm64, macOS, HTTP method/path egress rules, wake on request, live fork (no pause), agent harness API, hosted agent API (their own agent), gVisor or VM (no shared kernel)

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

## Caveats

- Plans carry extra non-spec keys pool_vcpu/pool_ram_gib/pool_disk_gib: Daytona concurrency is an org-wide resource pool per region, not a sandbox count (concurrency left null).
- Linux VM class is billed at the same meter as containers (billing docs; no separate VM line on the pricing page). 2026-09-28: re-split into two modes with identical rates because features differ: on-demand = container (no pause/fork, fs-only snapshots, no nested virt), vm-tier = Linux VM (pause, fork, memory snapshots, /dev/kvm). Card-level features describe the best available class; mode.features carry the per-class truth.
- Windows $0.0858/vCPU-h encoded as replacing the Linux vCPU rate; if additive, Windows vCPU would cost $0.1362/h.
- GPU: vCPU/RAM/disk billed on top of GPU rate per third-party tracker, not stated in official docs. Free credits not usable on GPU.
- Snapshot storage 'remains billed' but rate unpublished (null). Warm-pool idle billing undocumented.
- Egress/ingress pricing not published; treated as unknown (null). No public IPv4 offered.
- Billed transitions (Creating/Starting/Stopping/Pausing) have unknown duration (boot_overhead_s null).
- Idle tail billed: containers keep billing until 15-min idle auto-stop, VMs until 60-min auto-pause; background processes do not count as activity.
- Stopped/paused disk billed indefinitely except archived containers (auto-archive after 7 d stopped by default). 5 free GiB assumed per sandbox (unverified).
- Tier 1-2 have restricted outbound internet; full internet requires Tier 3 ($500 prepaid top-up).
- Per-sandbox default max 4 vCPU / 8 GiB / 10 GiB disk (raisable per region); disk can never shrink.
- $200 free credit is one-time; expiry term exists but unpublished. Startup credits: pricing page says up to $50k, startups page up to $100k.
- Volume discounts and Enterprise/dedicated/BYOC pricing unpublished.
- Verifier 2026-09-28: the engine ignores pool_vcpu/pool_ram_gib/pool_disk_gib and all tiers have concurrency null, so it always picks a $0 tier. A fleet needing >250 vCPU / >500 GiB concurrently really requires Tier 4 ($2,000 top-up every 30 days, a de facto usage floor), and Tier 1-2 cannot reach the open internet; neither is enforced by the estimate.
- Creation rate limits per tier (300/400/500/600 per min, https://www.daytona.io/docs/en/limits) are not modelled.
- Verifier 2026-09-28: https://www.daytona.io/pricing (static render) shows only the preemptible GPU column (matching the spot mode); on-demand GPU prices are primary-sourced from a Chrome render of the pricing page's On-demand tab (2026-09-28, research/raw/daytona-On-demand.txt: B300 $7.10, B200 $6.25, MI355X $5.99, H200 $4.54, H100 $3.95); the static fetch shows only the preemptible column. The pricing page lists Windows as a separate 'OS, Windows $0.0858/vCPU/h' line, which reads like an additive surcharge (reading B); the card keeps reading A (replacement) - unresolved.
- perf is mode-level (card-level nulled): on-demand (container) carries ComputeSDK TTI/DAX; vm-tier (Linux VM) carries the HPC CPU result (18.6 runs/s). See research/perf-map.json.
- Open source: the public daytonaio/daytona repository states that as of June 2026 core development moved to a private codebase and the repo receives no further updates, fixes or releases. Historical v0.190.0 code remains AGPLv3; it is not the current managed service. Sources: https://github.com/daytonaio/daytona/blob/ec4c21b2d597091ac09ecc278f3bcc172575a987/README.md , https://github.com/daytonaio/daytona/blob/v0.190.0/LICENSE
- 2026-09-28 (docs/limits): org-wide pools Tier 1 10 vCPU / 20 GiB / 30 GiB (was recorded as 10 GiB RAM), Tier 2 100/200/300, Tier 3 250/500/2,000, Tier 4 500/1,000/5,000; creates/min 300/400/500/600. Encoded as plan max_total_vcpu / max_total_ram_gib (not yet enforced by the engine) and max_starts_per_min (enforced). Snapshot storage rate still unpublished (null).
- 2026-09-28 features: nested virtualization (/dev/kvm) only on Linux VM sandboxes created with kvm=true, and only if the org has the sandbox_kvm feature (enablement path undocumented). Container, GPU and Windows classes: no nested virt. Desktop: VNC/xfce4 desktop is in the default container image and Linux VM/Windows run at a fixed VNC resolution; custom images need the packages installed; GPU class desktop unknown (null). Docker-in-Docker documented for sandboxes (docker:dind snapshots, Docker Compose).
- Wall-clock TTL (v0.199.0, 2026-07-17): optional per-sandbox ttl_minutes destroys the sandbox in any state after a fixed deadline. It is disabled by default, so max_session_h stays null. It serves as a per-sandbox runaway-cost guard, not an org-level cap.
- Top-ups are prepaid wallet credit and non-refundable. The Tier 2/3/4 thresholds ($25, $500, $2,000 every 30 days) are therefore sunk cash if the usage is not consumed, and a negative balance can suspend the org. Expiry and rollover of paid credit are not published. SLA service credits apply only to Tier 4 and above.
- Rate limits per tier (docs/limits): general API 10k/20k/40k/50k per min; sandbox creation 300/400/500/600 per min; lifecycle ops (start/stop/etc.) 10k/20k/40k/50k per min. max_starts_per_min holds the creation limit, so restarting existing sandboxes is far less constrained.

## How this provider charges


Daytona has **no plan fees**. One price sheet ($0.0504/vCPU-h, $0.0162/GiB-h, $0.000108/GiB-h disk after 5 GiB free, per second, on **reserved** resources) applies to every sandbox class. What changes the bill is:
- the sandbox class (container, Linux VM, Windows, GPU on-demand, GPU spot),
- the lifecycle state,
- the prepaid-top-up tier, which sets the org-wide resource pool and network access.

## Regime table

| # | Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|---|
| 1 | Container sandbox, on-demand (default class) | Default `daytona-*` snapshots or any OCI image | Per second on **reserved** vCPU + RAM + disk while Started. Utilisation is not considered | $0.0504/vCPU-h, $0.0162/GiB-h. 4 vCPU/8 GiB = **$0.3312/h**. Default 1/1/3; per-sandbox max 4 vCPU/8 GiB/10 GiB (org limit, raisable per region) | https://www.daytona.io/pricing, https://www.daytona.io/docs/en/billing |
| 2 | Linux VM sandbox | `daytona-vm-*` class (fork, pause/resume, memory snapshots, nested KVM) | Same billing table as containers. No separate VM price is published | Assumed same $0.0504/$0.0162 (no VM line on the pricing page) | https://www.daytona.io/docs/en/sandboxes.md, /docs/en/billing |
| 3 | Windows VM sandbox | `windows-small/medium/large` presets only | Windows vCPU rate. RAM and disk are presumably the standard rates | $0.0858/vCPU-h. Presets 1/4/30, 2/8/50, 4/16/50 (vCPU/GiB/GiB disk). It is **ambiguous** whether $0.0858 replaces $0.0504 (A) or is added to it (B). windows-large: A = $0.6024/h, B = $0.8040/h (compute only) | https://www.daytona.io/pricing, /docs/en/sandboxes.md |
| 4 | GPU sandbox, on-demand | GPU class (container with exclusive GPUs, up to 8 GPUs) | GPU $/h per GPU. Per a third-party tracker (not in the official docs), vCPU/RAM/disk are metered on top at the normal rates. Deleted on stop | B300 6.25, B200 6.25, MI355X 5.99, H200 4.54, H100 3.95, RTX PRO 6000 3.03, RTX 5090 1.29, RTX 4090 0.99 ($/GPU-h). Each GPU unlocks up to 16 vCPU / 192 GB / 512 GB | https://www.daytona.io/pricing, https://gpuadvisor.com/providers/daytona |
| 5 | GPU sandbox, **spot/preemptible** (added v0.205.0, 2026-08-18) | `spot` flag on GPU create | Same as 4 at a lower GPU rate. Can be evicted **without notice** when on-demand needs capacity (`spotEvictedAt`). Does **not** count against GPU quota | B300 4.08, B200 3.59, MI355X 3.44, H200 2.61, H100 2.27, RTX PRO 6000 1.74, RTX 5090 0.74, RTX 4090 0.57 (about 35-43% off) | https://www.daytona.io/pricing, /docs/en/sandboxes.md, https://www.daytona.io/changelog |
| 6 | Transitional states | Creating, Starting, Stopping, Pausing | Billed **like Started** (full vCPU+RAM+disk). The reduced rate only begins once the Stopped/Paused state is reached | Duration not published | /docs/en/billing |
| 7 | Idle-but-running (auto-stop tail) | Until the idle timer fires | Full running rate. Idle = no *external* activity. Background processes do not reset the timer | Container/GPU auto-stop 15 min default. VM auto-pause 60 min default. `0` = never stop (bills forever) | /docs/en/sandboxes.md |
| 8 | Stopped | Container or Linux VM stopped | **Disk only** | $0.000108/GiB-h above 5 GiB = $0.0788/GiB-month. Stopped containers still occupy disk *quota* until archived. Stopped VMs offload state and free the quota but are still billed | /docs/en/billing, /docs/en/limits |
| 9 | Paused (VM / Windows only) | Pause (memory preserved) | **Disk only**. Preserved memory state is **not billed** | Same disk rate | /docs/en/billing |
| 10 | Archived (containers only) | Manual, or auto-archive after 7 days stopped by default (max 30) | **$0** (filesystem moved to object storage, quota freed) | 0 | /docs/en/billing, /docs/en/limits |
| 11 | Ephemeral | `autoDeleteInterval=0`. GPU sandboxes are always ephemeral | Deleted on stop, so no stopped-disk billing | 0 after stop | /docs/en/sandboxes.md |
| 12 | Snapshots created from a sandbox (cold fs, or hot fs+memory for VMs) | Retained snapshots | "Remain billed for storage" even after the sandbox is deleted. **Rate not published**. They deactivate after 2 weeks unused. Default quota 30 snapshots (v0.175.0) | null | /docs/en/billing, /docs/en/snapshots.md, /changelog |
| 13 | Volumes (S3-backed FUSE) | Shared data | **Free**, and not counted in the storage quota | $0. Max 100 volumes/org | /docs/en/volumes.md |
| 14 | Warm pools | Pre-created running sandboxes per snapshot/region (gated, contact support) | Count against quota like running sandboxes. **Billing while idle in the pool is not documented** (likely running rate) | null | /docs/en/warm-pools.md |
| 15 | Tier 1 (email verified) | Default | No fee. Org pool 10 vCPU / 10 GiB / 30 GiB (limits page re-checked 2026-09-28; an earlier "20 GiB" reading no longer appears). 300 creates/min. **Restricted egress** (allowlisted essentials only) | $0 | /docs/en/limits |
| 16 | Tier 2 | Card linked + $25 top-up (prepaid credit) | Pool 100 vCPU / 200 GiB / 300 GiB. 400 creates/min. Egress still restricted | $25 one-time prepaid | /docs/en/limits, /docs/en/network-limits.md |
| 17 | Tier 3 | $500 top-up (prepaid credit) | Pool 250 vCPU / 500 GiB / 2 TB. 500 creates/min. **Full internet** | $500 one-time prepaid | /docs/en/limits |
| 18 | Tier 4 | **$2,000 top-up every 30 days** | Pool 500 vCPU / 1000 GiB / 5 TB. 600 creates/min. In effect a $2k/month minimum spend, consumed as credit | $2,000 per 30 days | /docs/en/limits |
| 19 | Enterprise / dedicated region / custom region (BYOC) | Contact sales | Custom limits, SSO/SCIM, audit logs. Dedicated = Daytona-managed single-tenant. Custom = your runners, "unlimited" bounded by your compute. **Prices unpublished** | null | https://www.daytona.io/pricing, /docs/en/regions.md |
| 20 | Volume discounts | "as you scale" | Unpublished | null | https://www.daytona.io/pricing |
| 21 | Free credit | Sign-up, no card | $200 one-time. Consumed before paid credit. **Not usable on GPU**. Expiry exists ("See breakdown and expiration") but the term is unpublished | $200 | /pricing, /docs/en/billing |
| 22 | Startup program | Application | Pricing page says "up to $50k". Startups page says "up to $100k, $10K straight away". Inconsistent | $10k-$100k | https://www.daytona.io/startups |
| 23 | Regions | `us`, `eu` shared. GPU runs in virtual `earth` region (region preference ignored) | **No regional price differences published** | multiplier 1 | /docs/en/regions.md |
| 24 | Egress / ingress / IP | Any | **Not published and not metered on the pricing page**. No public IPv4 offered (HTTPS preview proxy + SSH gateway only) | null | /docs/en/network-limits.md |

## Gotchas

1. **Utilisation doesn't matter.** Billing is on *reserved* vCPU/RAM. At 30% CPU util you still pay 100%. Resize live (up only while running) to right-size.
2. **The idle tail is billed at full rate**, and "idle" ignores your own background processes. A container keeps billing 15 min after the last external call. A VM bills a full **60 min** before auto-pause. Setting auto-stop `0` means it runs and bills until you stop it. There is no org-level max-runtime guard (open feature request, GitHub daytonaio/daytona#4158).
3. **Stopping ≠ free.** Stopped and paused sandboxes bill disk (above 5 GiB) indefinitely. Only *archived* containers or deleted sandboxes are free. VMs cannot be archived, so a paused/stopped VM pays disk forever.
4. **Stop/pause transitions bill as running**, and so do Creating/Starting.
5. **Snapshots keep billing after the sandbox is deleted**, and the rate is unpublished.
6. **The tier system is a hidden gate.** Tiers 1-2 have **restricted outbound internet** (package registries, git hosts, model APIs only). Full internet needs a $500 top-up (Tier 3). More than 250 vCPU concurrent needs the recurring $2,000/30-day top-up (Tier 4). Concurrency is an org-wide vCPU/RAM/disk *pool*, not a sandbox count.
7. **Disk quota and disk billing are decoupled.** Stopped containers eat quota (which can block creates even when CPU is free) until archived. Stopped VMs free the quota but still bill.
8. **Windows** comes only as presets with 4 GiB RAM per vCPU. A 4 vCPU/8 GiB need forces windows-large (16 GiB, 50 GiB disk), and the vCPU rate is 1.7x Linux (or 2.7x if additive). Windows sandboxes also count 16 GiB each against the RAM pool.
9. **GPU:** the $200 free credit cannot be spent on GPUs. GPU sandboxes are deleted on stop, so no pause and no persistence. Spot GPUs are killed without warning. GPU work on shared regions ignores the region choice ("earth").
10. **Free-credit reassignment.** Redeeming free credits later in the month retroactively re-assigns usage and releases paid balance. Billing lags up to 48 h, which matters for spend monitoring. No spend-cap feature is documented. Auto top-up can keep charging the card.
11. Per-sandbox default max is **4 vCPU / 8 GiB / 10 GiB disk** unless raised per region. Disk can never shrink.

## Worked example

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

Assumptions:
- Container/VM disk = 10 GiB (daytona-large). Billable disk = 5 GiB × $0.000108 = $0.00054/h per sandbox.
- The 5 GiB free allowance is read as per sandbox (unverified).
- Persistent sandboxes are stopped outside hours: 730 − 176 = 554 h × 50 = 27,700 stopped sandbox-hours. That is below the 7-day auto-archive, so they are never archived.
- Tier: 50 × 4 vCPU = 200 vCPU and 400 GiB RAM needs **Tier 3** (one-time $500 top-up, spent as credit, so no extra cost). Windows needs 800 GiB RAM, so **Tier 4** ($2,000/30-day top-up, exceeded by usage anyway).

| Regime | Compute (8,800 h) | Disk running | Disk stopped/paused | Snapshots 50 GiB | Egress 100 GiB | Plan fee | **Monthly total** |
|---|---|---|---|---|---|---|---|
| Container, persistent (stop nightly) | 8,800 × 0.3312 = $2,914.56 | $4.75 | $14.96 | unpublished (null)* | not published ($0 assumed) | $0 (Tier 3) | **$2,934.27** + snapshots |
| Container, ephemeral (delete each day) or archived nightly | $2,914.56 | $4.75 | $0 | null* | $0? | $0 | **$2,919.31** + snapshots |
| Linux VM, paused/stopped nightly (price assumed same as container) | $2,914.56 | $4.75 | $14.96 | null* | $0? | $0 | **$2,934.27** + snapshots |
| Windows windows-large (4/16/50), reading A: $0.0858 replaces $0.0504 | 8,800 × 0.6024 = $5,301.12 | 45 GiB → $42.77 | $134.62 | null* | $0? | $0 (Tier 4 top-up consumed) | **$5,478.51** |
| Windows, reading B: $0.0858 added to $0.0504 | 8,800 × 0.8040 = $7,075.20 | $42.77 | $134.62 | null* | $0? | $0 | **$7,252.59** |
| GPU on-demand / spot | n/a (workload has no GPU). Adding 1 H100 each would add 8,800 × 3.95 = $34,760 (on-demand) or 8,800 × 2.27 = $19,976 (spot) | | | | | | |

\* Snapshot rate unpublished. *If* charged at the disk rate ($0.0788/GiB-month), 50 GiB would cost $3.94. This is illustrative only.

Other effects on the total:
- The **30% CPU utilisation does not reduce any figure**, because billing is on allocation.
- **Idle-tail sensitivity:** if each sandbox is left to the idle timer instead of being stopped explicitly:
  - containers add 15 min × 50 × 22 = 275 h, or **+$91.08**
  - VMs add 60 min × 50 × 22 = 1,100 h, or **+$364.32**
- The $200 one-time free credit reduces month 1 only (or about $16.67/month amortised over 12 months).

Sources: https://www.daytona.io/pricing · https://www.daytona.io/docs/en/billing · https://www.daytona.io/docs/en/limits · https://www.daytona.io/docs/en/sandboxes.md · https://www.daytona.io/docs/en/snapshots.md · https://www.daytona.io/docs/en/volumes.md · https://www.daytona.io/docs/en/warm-pools.md · https://www.daytona.io/docs/en/regions.md · https://www.daytona.io/docs/en/network-limits.md · https://www.daytona.io/changelog · https://www.daytona.io/startups · https://github.com/daytonaio/daytona/issues/4158 · https://gpuadvisor.com/providers/daytona (third-party, GPU billing composition)