# Exoscale

Official pricing: https://www.exoscale.com/pricing/  
Category: hyperscaler · Isolation: vm

## Pricing regimes (raw)

- **Standard** (sizes): standard-micro 1 vCPU/0.5 GiB $0.00729/h; standard-tiny 1 vCPU/1 GiB $0.01458/h; standard-small 2 vCPU/2 GiB $0.02333/h; standard-medium 2 vCPU/4 GiB $0.04666/h; standard-large 4 vCPU/8 GiB $0.09333/h; standard-extra_large 4 vCPU/16 GiB $0.18667/h
- **Cpu** (sizes): cpu-extra_large 8 vCPU/16 GiB $0.23333/h; cpu-huge 16 vCPU/32 GiB $0.46666/h; cpu-mega 32 vCPU/64 GiB $0.93333/h; cpu-titan 40 vCPU/128 GiB $1.86667/h
- **Memory** (sizes): memory-extra_large 2 vCPU/16 GiB $0.14/h; memory-huge 4 vCPU/32 GiB $0.28/h; memory-mega 8 vCPU/64 GiB $0.56/h; memory-titan 12 vCPU/128 GiB $1.12/h
- **Storage optimized** (sizes): storage-extra_large 4 vCPU/16 GiB $0.14/h; storage-huge 8 vCPU/32 GiB $0.28/h; storage-mega 12 vCPU/64 GiB $0.56/h; storage-titan 16 vCPU/128 GiB $1.12/h; storage-jumbo 24 vCPU/225 GiB $1.86667/h
- **GPU A30** (sizes): gpua30-small 12 vCPU/56 GiB $0/h
- **GPU2 V100** (sizes): gpu2-small 12 vCPU/56 GiB $0/h
- **GPU3 A40** (sizes): gpu3-small 12 vCPU/56 GiB $0/h
- **GPU A5000** (sizes): gpua5000-small 12 vCPU/56 GiB $0/h

## Features

Yes: snapshots, fork/clone, persistent disk, volumes, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, full VM (own kernel), Docker inside, root, systemd, SSH, public IPv4, custom domains, raw TCP inbound, egress allowlist, open internet, static egress IP, private networking, SOC 2, HIPAA, SSO, GPU, Windows, live fork (no pause), EU data residency, snapshots on demand, gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), extra volumes

No: memory snapshots, pause/resume, 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, preinstalled agents, HTTPS preview URLs, self-hosting, BYOC, open source, arm64, macOS, Python, Node.js, GPU desktop, HTTP method/path egress rules, wake on request, live resize, memory fork, MCP server, automatic snapshots, 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/exoscale.json → feature_evidence

## Caveats

- DIY infrastructure rather than an agent-managed sandbox: guest installation, HTTPS/domain setup, abuse controls and operations are customer responsibilities. Linux guest administrative capabilities are not preinstalled agent features. 30% guest CPU utilization does not reduce allocated-compute charges. Unknowns remain null; a partial subtotal is not a complete invoice.
- Large 4/8 compute $0.09333/hour; 50GB root disk adds $0.007/hour. Root SSD is not included despite table minimum-storage column.
- 1.42GB outbound per running instance-hour pooled within organization, reset/accounted hourly; overage $0.02/GB. Static card allowance set to 0 conservatively; it overstates egress for the example, which is within aggregate allowance if distributed normally.
- Primary management IP included; extra Elastic IP $0.01389/hour (~$10.1397/730h); no global free extra IPv4 assumption.
- Deleting an instance deletes its snapshots; export or custom-template workflow needed for durable delete/recreate. Snapshot operations may take hours; 30 snapshots/account default, not nameable; storage-optimized snapshots unsupported.
- Card represents compute+storage floor. Account transfer timing and disk retention must be modeled separately.
- Boot time not published (community docs only say "Launch virtual machines within seconds") and not in benchmarks.json; boot_overhead_s left unset (2026-09-28).
- Egress free allowance is 1.42 GiB per running instance-hour pooled across the organization (~1 TiB/month per always-on VM); card egress_free_gib=0 overstates egress cost for any workload with running VMs.
- Trial credit: new accounts "might be entitled" to a one-time 20 EUR/CHF credit after saving a card (post-paid mode); Exoscale may withhold or expire it. Not modelled (card is USD, no stated FX).
- Windows licence (not modelled; modes are Linux): $0.01111/h per VM up to 7 vCPU, $0.02222/h from 8 vCPU, billed from creation to deletion including while stopped (portal.exoscale.com/api/pricing/licenses).

## How this provider charges


DIY infrastructure rather than an agent-managed sandbox: guest installation, HTTPS/domain setup, abuse controls and operations are customer responsibilities. Linux guest administrative capabilities are not preinstalled agent features. 30% guest CPU utilization does not reduce allocated-compute charges. Unknowns remain null; a partial subtotal is not a complete invoice.

Prices below are public list prices unless a row explicitly says promotion/commitment. USD; taxes excluded unless noted.

| Regime | When / billing | Captured prices | Source |
|---|---|---|---|
| Standard | On demand; USD list compute only; local root SSD minimum 10 GB and priced separately. No monthly cap; compute meter stops powered off. | standard-large: $0.09333/h; cap None | https://www.exoscale.com/pricing/ |
| Cpu | On demand; USD list compute only; local root SSD minimum 10 GB and priced separately. No monthly cap; compute meter stops powered off. | cpu-extra_large: $0.23333/h; cap None | https://www.exoscale.com/pricing/ |
| Memory | On demand; USD list compute only; local root SSD minimum 10 GB and priced separately. No monthly cap; compute meter stops powered off. | memory-huge: $0.28/h; cap None | https://www.exoscale.com/pricing/ |

## Billing details
- **granularity_seconds**: 1
- **minimum_billed_seconds**: Unknown; not assumed zero.
- **setup_fee_usd**: 0
- **api**: True
- **cli**: True
- **ipv 6**: True
- **ingress_usd**: 0
- **cpu**: KVM shared physical infrastructure; CPU optimized is a shape ratio, not a verified dedicated-core guarantee.
- **minimum_term**: Per-second usage, no long-term commitment.
- **backup**: Snapshots and local SSD $0.00014/GB-hour = $0.1022/GB per 730h. Custom templates $0.00028/GB-hour.
- **provisioning**: Unknown; not assumed zero.

## Gotchas
- DIY infrastructure rather than an agent-managed sandbox: guest installation, HTTPS/domain setup, abuse controls and operations are customer responsibilities. Linux guest administrative capabilities are not preinstalled agent features. 30% guest CPU utilization does not reduce allocated-compute charges. Unknowns remain null; a partial subtotal is not a complete invoice.
- Large 4/8 compute $0.09333/hour; 50 GB root disk adds $0.007/hour. Root SSD is not included despite table minimum-storage column.
- 1.42 GB outbound per running instance-hour pooled within organization, reset/accounted hourly; overage $0.02/GB. Static card allowance set to 0 conservatively; it overstates egress for the example, which is within aggregate allowance if distributed normally.
- Primary management IP included; extra Elastic IP $0.01389/hour (~$10.1397/730h); no global free extra IPv 4 assumption.
- Deleting an instance deletes its snapshots; export or custom-template workflow needed for durable delete/recreate. Snapshot operations may take hours; 30 snapshots/account default, not nameable; storage-optimized snapshots unsupported.
- Card represents compute+storage floor. Account transfer timing and disk retention must be modeled separately.

## Worked example
4 vCPU / 8 GiB; 50 concurrent × 8 h/day × 22 days = **8,800 guest-hours**; 30% CPU busy; **50 GiB-month total snapshots**, **100 GiB total egress**. 50 GiB retained state is total, not 50 GiB per VM. Compute-only subtotals below exclude unknown components and tax.

| Mode | Delete/recreate / active-window compute | 50 retained hosts, 730-hour month |
|---|---:|---:|
| Standard | $821.30 | $3,406.55 |
| Cpu | $2,053.30 | $8,516.55 |
| Memory | $2,464.00 | $10,220.00 |

Snapshot subtotal: $5.1100; rate $0.10219999999999999/GiB-month. Egress: see allowances, pooling/accrual and rate $0.02/GiB; not every mode has the same allowance.
Powered-off compute billed: False. A retained-host column is the always-running upper scenario when this is false; powered-off storage/IP still accrue.
Monthly caps attach to instance identities, not to a pool of replacement instances. The active-window column requires deleting/releasing billable compute between shifts for products that bill while off. Restoring from disk is not memory resume. The engine cannot perfectly model these lifecycles; see validation/estimator-gaps.md.

## Sources
- https://www.exoscale.com/pricing/
- https://www.exoscale.com/compute/
- https://portal.exoscale.com/api/pricing/opencompute
- https://community.exoscale.com/platform/billing/
- https://community.exoscale.com/product/compute/instances/overview/
- https://community.exoscale.com/product/compute/instances/how-to/snapshot/