# exe.dev

Official pricing: https://exe.dev/pricing  
Category: dev-env · Isolation: vm

## Pricing regimes (raw)

- **Work pool (current; reserved pool billed hourly 24/7, shared by all VMs)** (pool)
- **Personal pool (current; 2/4/8/16 vCPU at $15/$35/$75/$155)** (pool)
- **Standalone / sandbox VMs (current; per VM, own reserved capacity)** (resource): $0.0525/vCPU-h, $0/GiB-h
- **Usage / Cloud Pool (sales-only, per second, peak-hourly CPU + active memory)** (resource): $0.05/vCPU-h, $0.016/GiB-h [sales]

## Features

Yes: fork/clone, pause/resume, persistent disk, ≥24 h sessions, custom image (Docker or snapshot), your own Docker/OCI image, full VM (own kernel), Docker inside, nested virtualization, root, systemd, browser, preinstalled agents, SSH, HTTPS preview URLs, custom domains, open internet, private networking, BYOC, SSO, Python, Go, credential injection, live resize, live fork (no pause), EU data residency, scoped API keys, MCP server, agent harness API, gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), secret proxy, secret proxy for any API

No: snapshots, memory snapshots, volumes, idle auto-stop, start from your own snapshot, desktop GUI, computer-use API, code interpreter, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, public IPv4, raw TCP inbound, egress allowlist, static egress IP, self-hosting, open source, SOC 2, HIPAA, GPU, arm64, Windows, macOS, Node.js, GPU desktop, HTTP method/path egress rules, wake on request, memory fork, webhooks, automatic snapshots, snapshots on demand, hosted agent API (their own agent), extra volumes, shared volumes

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

## Caveats

- Three conflicting price surfaces. Live /pricing and /docs/pricing (2026-09-28) show the NEW model: Personal $15 (2 vCPU/4 GB), Work $0.21/h per 4 vCPU/8 GB (min $150), standalone VMs $0.105/2 vCPU-h. docs/billing/overview.md and the /sandbox page still show LEGACY $20/$40/$80/$160 pools (documented as current as late as 2026-09-09). The /sandbox usage rates ($0.05/core-h, $0.016/GiB-h) are sales-only 'Cloud Pool'.
- Work pool prices confirmed linear from the embedded pricing JSON (4/8/16/32/64/128/256/512 vCPU = $0.21/$0.42/$0.84/$1.68/$3.36/$6.72/$13.44/$26.88 per h); intermediate even sizes (CLI allows any even 4..512) interpolated at the same $0.0525/vCPU-h.
- Personal pool sizes 2/4/8/16 vCPU = $15/$35/$75/$155 per month (embedded pricing JSON on exe.dev/pricing, verified 2026-09-28). Note 4 vCPU/8 GB is $35, not 2x$15.
- Pools bill regardless of utilisation; engine should size by peak concurrent vCPU/RAM. Resizing pools off-hours can reduce cost (not modelled; hourly proration assumed but undocumented).
- Standalone VM billing granularity, minimum, and whether stopped standalone VMs bill are undocumented; RAM is bundled (default 4 GiB/vCPU, max 2 GiB/vCPU at top size).
- Usage pricing bills CPU on PEAK usage within each hour, so 'active' basis understates cost for bursty loads; memory billed on 'active' memory (definition unclear). Minimum commitment unknown.
- included_disk_gib (100) is POOLED per account/user (Work: 100 GB per 2 vCPU of pool), not per sandbox. Disk measured as time-averaged ext4 filesystem usage, not provisioned size. Default VM disk 25 GB, grow-only.
- No snapshots: retained state = stopped VM disk (or cp clone), billed as disk at $0.08/GB-month beyond allowance.
- Egress allowance: 200 GB (Personal, any size; legacy Individual), 250 GB per user (legacy Team), Work 50 GB per vCPU of pool (200 GB at 4 vCPU, capped at 800 GB) - the '200 GB' bullet is the 4-vCPU value, not a conflict. Overage $0.05/GiB billed at cycle end; inbound never billed.
- Standalone VMs and pools require support/team approval.
- No free compute tier; 30-day trial via invite; invite rewards (credits/memory/disk) amounts unpublished. Legacy plans include $20/mo Shelley LLM credit (not compute).
- No dedicated IPv4 (shared per-owner IPv4 pool; HTTPS proxy + SSH only). Account bound to one region (Personal); Work pools in any region; no regional price differences found.
- AWS/Azure Marketplace billing exists (legacy sizes); marketplace prices not retrieved.
- No benchmark data for exe-dev in research/benchmarks.json.
- Usage-based (sales-only) mode: CPU and active memory bill on the PEAK of each hour, so set the peak CPU/RAM inputs, not the averages. "Active memory" is not defined (page cache inclusion unknown); the idle trigger that drops a VM to disk-only billing is not documented.
- Usage-priced Cloud Pool ($0.05/core-h peak, $0.016/GiB-h active memory, exe.dev/sandbox) is not self-serve: the billing docs say to email support@exe.dev, so the mode keeps the sales flag.
- Pricing changed 2026-09-28 (blog 'Pools: Our Newest Primitive'): Personal/Work replaced Individual/Team; existing customers have one year to migrate. Personal/Work/Enterprise include no monthly Shelley credit.
- A pool pending support approval is not billed; pool billing stops when the pool is deleted; pools reserve CPU+memory only, not disk.
- Switching Personal -> Work cancels Personal immediately with no automatic credit for the rest of the cycle (refund on request).
- Approval (docs 2026-09-29): Personal standalone VMs need no approval; Work standalone VMs need a team-admin setting (off by default); pools up to 32 vCPU are immediate, 34-512 vCPU need exe.dev support approval. Supersedes the older 'Standalone VMs and pools require support/team approval' caveat.

## How this provider charges


## Which price source is current?

exe.dev currently shows three price surfaces that disagree:

| Surface | What it says | Status (2026-09-28) |
|---|---|---|
| `exe.dev/pricing` and `exe.dev/docs/pricing` (both now serve the same page) | **Personal $15/mo** (2 vCPU / 4 GB pool, selector 2/4/8/16 vCPU), **Work $0.21/h** per 4 vCPU / 8 GB pool (selector 4..512, min spend $150/mo), Enterprise custom; **standalone VMs $0.105 per 2 vCPU per hour** | **CURRENT.** Live fetch on 2026-09-28 of `/pricing` returned this page. The earlier raw dump of `/pricing` (in `raw/exe_dev_pricing.txt`) still showed the $20/$25 tiers, so the switch happened recently. |
| `exe.dev/docs/billing/overview.md`, `/sandbox` pool box, Flavio Copes deep dive (2026-09-09), boxd comparison (May 2026) | Small/Medium/Large/XLarge pools: $20/$40/$80/$160 individual, $25/$50/$100/$200 per Team user; Enterprise "from $35.84/hr" (512 vCPU / 2048 GB) | **LEGACY.** It was still documented on 2026-09-09 and the billing docs have not been updated. Existing subscribers are probably grandfathered (unverified). Before that, from launch (Dec 2025) to about April 2026, the offer was "$20/month, 2 CPU / 8 GB / 25 GB disk, 25 VMs". |
| `/sandbox` "Usage" box | CPU $0.05/core-h, active memory $0.016/GiB-h, disk $0.08/GiB-mo; CPU and active memory billed on **peak hourly usage**, metered per second; idle VMs pay no CPU and memory at the disk rate | **Sales-only** ("Contact us about usage pricing"). This is the "Cloud Pool" plan in the billing docs ("100% usage based"). It is not self-serve. |

No blog post, changelog entry or dated announcement covers the switch. The blog index (through 2026-09-23) has only one billing post, about Shelley token auto-purchase on 2026-08-05. The CLI reference already matches the new model:
- `billing capacity --cpu=2|4|8|16` is the Personal selector.
- `pool new --cpus` takes an even number from 4 to 512, with "memory is 2 GiB per vCPU". These are Work pools.
- `new --standalone` / `--sandbox` gives "its own reserved capacity" and needs a plan with standalone VMs plus team approval.
- `team settings vm-placement poolless` is marked "(legacy)".

## Regimes

| # | Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|---|
| 1 | **Personal pool** (current) | Solo account | Flat monthly pool of vCPU and RAM shared by up to 50 VMs. Idle or stopped VMs free up pool capacity. No per-VM charge. | $15/mo for 2 vCPU / 4 GB; min spend $15. Selector 4/8 GB $35, 8/16 GB $75, 16/32 GB $155 (embedded pricing-scale JSON `personal:monthly:20260914` in the /pricing HTML; verified 2026-09-28 — corrected from "not published"). 100 GB disk (100 GB per 2 vCPU in the comparison table), 200 GB bandwidth, 1 pool, pool placed in the user's region. | exe.dev/pricing |
| 2 | **Work pool** (current) | Teams, any region | Reserved pool billed **per hour while it exists**, whatever the utilisation. Pools can be resized (`pool resize`) and there is no limit on the number of pools. | $0.21/h per 4 vCPU / 8 GB, i.e. $0.0525/vCPU-h with 2 GiB/vCPU included. That is $153.30/mo for a 4-vCPU pool left on 24/7. Min spend $150/mo. Up to 512 vCPU / 1,024 GB per pool. VMs per pool scale with size (embedded pricing JSON `work:monthly:20260901`): 4 vCPU = 100, 8 = 200, 16 = 400, 32 = 800, 64+ = 1,000 (the table's "1000 per pool" is the cap; `--max-vms` defaults to 100). Disk and bandwidth: 50 GB per vCPU (200 GB at 4 vCPU), capped at 800 GB — the "200 GB" bullets are the 4-vCPU values, not a conflict. Unlimited users and SSO. | exe.dev/pricing, docs/cli-pool.md |
| 2b | **Work pool resized to follow load** | You shrink the pool off-hours with `pool resize` | Same hourly rate on the current size. Hourly proration after a resize is **assumed**, not documented. | 4 vCPU floor | exe.dev/docs/cli-pool.md |
| 3 | **Standalone ("sandbox") VMs** (current) | `new --standalone` on Personal or Work. Needs team approval. Personal allows 50, Work 100. | Per VM on its own reserved capacity, priced **per vCPU only**. RAM is bundled: default 2 vCPU / 8 GB, max 16/32 on Personal and 32/64 on Work. Granularity and billing while stopped are **not documented** (the tagline says "by the second"). | $0.105 per 2 vCPU per hour = $0.0525/vCPU-h. That is $0.21/h at 4 vCPU. | exe.dev/pricing |
| 4 | **Legacy Individual pool tiers** | Subscribers from before about Sep 2026 (grandfathering unverified). Still in docs/billing. | Flat monthly pool, paid in advance, prorated on change. Overages billed at cycle end. | Small 2/8 $20 · Medium 4/16 $40 · Large 8/32 $80 · XLarge 16/64 $160. 50 VMs, 100 GB pooled disk, 200 GB transfer, $20/mo Shelley LLM credit. | docs/billing/overview.md |
| 5 | **Legacy Team per-user pools** | Legacy team subscriptions | Per seat. Each user has their own pool and can **burst into teammates' idle capacity**. | $25/$50/$100/$200 per user per month (Small..XLarge). 50 VMs per user, 100 GB disk, 250 GB transfer per user, SSO. | docs/billing/overview.md, /sandbox |
| 6 | **Legacy Enterprise / Reserved Cloud Pool** | Large fleets | Hourly reserved pool | "from $35.84/hr" for 512 vCPU / 2,048 GB ($0.07/vCPU-h with 4 GiB/vCPU). That is $26,163.20/mo if left on 24/7. "Thousands of VMs", AWS VPC. The current page says Enterprise is "Custom". | old /pricing, /sandbox |
| 7 | **Usage / Cloud Pool (sales)** | Negotiated with support@exe.dev | Per second. CPU on **peak CPU usage within each hour**, memory on **active** memory (peak hourly). Idle VMs pay no CPU and memory at the disk rate. | CPU $0.05/core-h, active RAM $0.016/GiB-h, disk $0.08/GiB-mo. Idle memory is $0.08/GiB-mo, about $0.00011/GiB-h. Included egress and any plan minimum are unknown. The rate may be negotiable. | exe.dev/sandbox, docs/billing/cloud-pool.md |
| 8 | **Hybrid** | "Pool your steady-state workers, burst overflow onto usage" | Sum of regimes 2 and 7 | n/a | exe.dev/sandbox |
| 9 | **Disk overage** | All plans | Time-averaged ext4 **filesystem usage** (not provisioned size) above the pooled allowance, in GiB-months. Stopped VMs keep paying disk. | $0.08/GB-month. Default VM disk is 25 GB and can only grow. | docs/billing/usage.md, faq/disk-usage |
| 10 | **Egress overage** | All plans | Running total of outbound traffic per cycle (per member on Team). Ingress is free. Billed at cycle end. | $0.05/GiB above the allowance | docs/billing/usage.md |
| 11 | **Shelley LLM credit** | Legacy plans (the current page lists "Shelley access") | Included LLM token credit at provider list price, no markup. Auto-purchase launched 2026-08-05. | $20/mo per user. It does **not** offset compute. | docs/billing/overview.md |
| 12 | **Trial / invite rewards** | New signups through an invite | 30-day trial. After the invitee upgrades, both sides get a reward (credits, extra memory or extra disk). | Amounts are not published | docs invites |
| 13 | **Cancellation** | After you cancel | Plan stays active to the end of the cycle, then drops to a free plan that can connect to existing VMs but not create new ones | $0 | docs/billing/subscriptions.md |
| 14 | **Marketplace billing** | AWS or Azure Marketplace subscription (`billing provider link --size=small..xlarge`) | Legacy sizes billed through the marketplace | Marketplace prices not retrieved (null) | CLI reference, AWS Marketplace listing |

Other facts:
- No region multipliers. The account is bound to one region, and PDX is closed to new accounts.
- No creation fee.
- No snapshots: `cp` is a copy-on-write disk clone, and retained state is simply disk.
- No dedicated IPv4: VMs share an IPv4 pool per owner, with HTTPS proxy and SSH only.
- No GPU.
- Nested virtualisation is Enterprise-only.

## Gotchas

- **The pool is flat. Your utilisation does not matter.** On a pool, 30% CPU utilisation saves you nothing, but idle or stopped VMs cost nothing extra, so 1,000 short sandboxes fit in one pool. You size the pool for **peak concurrent vCPU**. The legacy docs imply the sum of VM sizes is bounded by the pool ("run one VM at the full 16 vCPU, or split … 2×8 or 4×4"). Memory is soft-limited: exe.dev contacts you if active memory stays over the plan for a long time.
- **The current Personal plan is half the RAM of the legacy one**: 2 vCPU / 4 GB for $15, against 2 vCPU / 8 GB for $20.
- **Work costs ~5x Personal per vCPU at entry size**: $153.30 for 4 vCPU ($38.33/vCPU) vs $15 for 2 vCPU ($7.50/vCPU); at 16 vCPU it is $613.20 vs $155 (~4x). (Corrected: earlier text said "7.7x per vCPU", which compared a 4-vCPU pool to a 2-vCPU pool.) Personal cannot exceed 16 vCPU and 1 pool, which pushes fleets to Work or standalone VMs.
- **Standalone VMs cost the same per vCPU as Work pools** ($0.0525/vCPU-h). The difference is that they bill only while they exist (assumed), not 24/7, and they bundle up to 4 GiB/vCPU by default where a pool gives 2 GiB/vCPU.
- **Usage pricing bills CPU on the peak within each hour.** A sandbox that spikes to 100% for one minute in an hour pays a full core-hour for every core it used. The "30% util" saving is only real for flat loads.
- **Disk is billed on filesystem usage, time-averaged.** Short-lived sandboxes that write big temp data cost little. The allowance is pooled per account (per user on Team; per 2 vCPU on the new page), **not per VM**.
- **Apparent conflicts on the page mostly resolve via the embedded pricing JSON**: Work VMs per pool scale 100→1,000 with pool size; Work bandwidth/disk are 50 GB per vCPU (200 GB at 4 vCPU, cap 800 GB). Personal disk stays 100 GB at every size in the JSON, although the comparison table says "100 GB per 2 vCPU" (still a conflict).
- **Pools and standalone VMs need approval** ("reach out to support to ask for access" / "team approval").
- **Subscriptions are paid in advance and prorated**. Overages are billed in arrears on the renewal invoice.
- **There is no free compute tier.** The legacy $20 Shelley credit is for LLM tokens only.

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent sandboxes, 8 h/day, 22 days/month, 30% CPU utilisation, 50 GiB snapshots retained, 100 GiB egress.

That is 50 × 176 h = **8,800 sandbox-hours**. Peak demand is 200 vCPU / 400 GiB.

Storage and egress under the allowances:
- **Storage:** the 50 GiB "snapshot" is retained disk. It falls inside the 100 GB pooled allowance in every self-serve regime, so it costs $0. Under usage pricing it is 50 × $0.08 = $4.
- **Egress:** 100 GiB is inside the 200 GB-plus allowance in every self-serve regime, so it costs $0. Under usage pricing the allowance is unknown: $0 to $5.

| Regime | Calculation | Monthly total |
|---|---|---|
| 1 Personal pool | Maximum is 16 vCPU / 32 GB ($155) and 1 pool, so 200 vCPU / 400 GiB concurrent is **infeasible**. | n/a (infeasible) |
| 2 Work pool, 200 vCPU left on 24/7 | 50 × $0.21 × 730 h | **$7,665.00** |
| 2b Work pool resized to follow load | 50 × $0.21 × 176 h + 4-vCPU floor × $0.21 × 554 h = $1,848.00 + $116.34 (assumes hourly proration after `pool resize`) | **$1,964.34** |
| 3 Standalone VMs on Work (100-VM limit, fits) | 8,800 h × $0.21; exceeds the $150 minimum | **$1,848.00** (assumes deleted VMs stop billing and standalone usage counts toward the minimum) |
| 3 Standalone VMs on Personal (50-VM limit, just fits) | $15 pool + 8,800 × $0.21 | **$1,863.00** ($1,848 if the $15 is only a minimum spend) |
| 4 Legacy Individual pool | Maximum is XLarge 16/64, so this is **infeasible** | n/a |
| 5 Legacy Team per-user pools | 200 vCPU from 25 users × Large 8/32 at $100 = $2,500. The alternatives also come to $2,500 or more (50 × Medium $50, 100 × Small $25) or $2,600 (13 × XLarge $200). Workable only if sandboxes are spread over per-user pools or by bursting into teammates' capacity. | **$2,500.00** (requires 25 seats) |
| 6 Legacy Enterprise reserved pool | 512 vCPU at $35.84 × 730 h | $26,163.20 (oversized; it is "from", so smaller custom pools probably exist, price null) |
| 7 Usage pricing, CPU at a flat 30% | (4 × 0.3 × $0.05 + 8 × $0.016) = $0.188/h × 8,800 = $1,654.40, + $4 disk + $0–5 egress | **$1,658.40 – $1,663.40** (assumes all 8 GiB is "active" memory; plus any negotiated minimum, unknown) |
| 7 Usage pricing, bursty CPU (peak hour = 100%) | $0.328/h × 8,800 = $2,886.40, + $4 + $0–5 | **$2,890.40 – $2,895.40** |

The cheapest self-serve option is standalone VMs, at about $1,848. The sales-only usage plan is cheaper only if CPU is flat at about 30% within each hour.

Sources:
- https://exe.dev/pricing
- https://exe.dev/docs/pricing
- https://exe.dev/sandbox
- https://exe.dev/docs/billing/overview.md
- https://exe.dev/docs/billing/usage.md
- https://exe.dev/docs/billing/subscriptions.md
- https://exe.dev/docs/billing/cloud-pool.md
- https://exe.dev/docs/cli-pool.md
- https://exe.dev/llms-full.txt
- https://flaviocopes.com/exe-dev/
- https://boxd.sh/blog/boxd-vs-exe-dev/
- https://blog.exe.dev/