Direct answer: at Azure Chaos Studio public list prices on 11 November 2026 (USD, excluding tax and discounts), billing is per experiment action-minute: $0.10/action-minute in most commercial regions. One 30-minute action = $3; three 30-minute actions = $9; five 60-minute actions = $30; 20 experiments/month at 3 actions×30 minutes = $180/month; 50 experiments at 5 actions×60 minutes = $1,500/month. Chaos Studio itself is cheap; the real budget risk is the resources the experiment induces: autoscale VMs, database failovers, extra storage/network, monitoring logs and on-call labour. Workspaces/Scenarios are in public preview and Microsoft explicitly says they are not for production; use Experiments (classic) when you need the GA model.
This guide is for SRE, platform engineering, QA, security, finance and procurement running chaos engineering/failure drills, evaluating Chaos Studio versus self-hosted Chaos Mesh/Gremlin, or budgeting reliability testing. It separates official prices, editorial calculations and exclusions, covering action-minute billing, experiment-scenario TCO, Workspaces preview vs Experiments classic, induced-resource costs, selection and a 90-day checklist. Prices come from the Chaos Studio pricing page and Microsoft Learn overview/Workspaces docs (11 November 2026). This is not procurement advice.
(1) Billing is $0.10 per action-minute: one 30-minute fault = $3, three = $9; experiments × actions × duration is the base. (2) The expensive part is induced resources: autoscale, failover, extra VMs/storage/logs/on-call can dwarf $0.10/minute. (3) Workspaces/Scenarios are public preview and officially not for production; use GA Experiments (classic) for production drills, or keep Workspaces in non-production.
Billing: action-minutes and scope
The pricing page states Chaos Studio is pay-as-you-go per experiment execution, charged per action-minute or for the duration that experiment actions run; most commercial regions are $0.10/action-minute. An experiment has steps/branches/actions; an action can inject a fault or wait. This article calculates each action's actual run minutes; whether wait/other actions bill follows the actual invoice. Parallel branches do not reduce the sum: three parallel 30-minute actions are still 90 action-minutes. The page also warns that running chaos experiments may incur charges for other services (e.g., CPU triggering autoscale to add resources).
| Item | Price/basis | Note |
|---|---|---|
| Action-minute | $0.10/minute | Most commercial regions; per action run time |
| Parallel actions | Billed separately | Parallelism does not reduce action-minutes |
| Wait actions | Follow the bill | Official basis is action run time; this article assumes fault actions |
| Experiment resource | No separate fixed fee | Per action-minute, not per experiment |
| Induced resources | Separate | VM/DB/storage/network/logs, etc. |
Source: Chaos Studio pricing page (11 November 2026). USD, most commercial regions; excludes tax and discounts; wait/action counting follows the official bill.
Action-minute TCO: 1/3/5/10 actions
At $0.10/action-minute: 1 action×1 minute = $0.10; 1×30 = $3; 3×30 = $9; 5×60 = $30; 10×30 = $30; 1×24 hours = $144; 5×24 hours = $720. Action-minutes are linear; more actions/longer durations cost more, but compared with induced resources, $0.10/minute is rarely the main cost. Split the experiment into the faults you truly need, then decide duration/parallelism.
| Scenario | Action-minutes | Cost |
|---|---|---|
| 1 action × 1 min | 1 | $0.10 |
| 1 action × 30 min | 30 | $3.00 |
| 3 actions × 30 min | 90 | $9.00 |
| 5 actions × 60 min | 300 | $30.00 |
| 10 actions × 30 min | 300 | $30.00 |
| 1 action × 24 h | 1,440 | $144.00 |
| 5 actions × 24 h | 7,200 | $720.00 |
Editorial calculation: actions × minutes × $0.10; parallel actions bill individually; excludes wait/other actions, induced resources, tax and discounts.
Monthly experiments: 5/20/50/100
By month: 5 runs×3 actions×30 min = 450 action-minutes = $45; 20×3×30 = 1,800 = $180; 50×5×60 = 15,000 = $1,500; 100×10×30 = 30,000 = $3,000. If each run also includes wait/verification actions, action-minutes rise; if you only run one big drill per month, the base is small and the budget should go to drill environments, induced resources and labour. Running chaos on every CI pipeline multiplies frequency×actions quickly.
| Monthly experiments | Calculation | Action-minutes | Cost/month |
|---|---|---|---|
| 5 × 3 actions × 30 min | 5×3×30 | 450 | $45.00 |
| 20 × 3 actions × 30 min | 20×3×30 | 1,800 | $180.00 |
| 50 × 5 actions × 60 min | 50×5×60 | 15,000 | $1,500.00 |
| 100 × 10 actions × 30 min | 100×10×30 | 30,000 | $3,000.00 |
| 20 × 5 actions × 24 h | 20×5×1,440 | 144,000 | $14,400.00 |
Editorial calculation: runs×actions×minutes×$0.10; excludes wait/other actions, induced resources, tax and discounts; control duration carefully for long drills.
Workspaces (preview) vs Experiments (classic)
The current resource model is Chaos Studio Workspaces + Scenarios: discover resources by scope, choose a scenario, run it, review reports. But Workspaces/Scenarios are in public preview; Microsoft explicitly says no SLA/limited warranty, not meant for production, best-effort support. Experiments (classic) is the GA legacy model using experiment/target/capability/fault, supporting service-direct (faults against Azure resources) and agent-based (CPU/memory/process faults inside VMs/VMSS). Use classic when you need production/GA; Workspaces fits non-production trials/scenario catalog. Both bill per action-minute, but the model/features/support boundary differs.
| Item | Workspaces/Scenarios | Experiments (classic) |
|---|---|---|
| Status | Public preview | GA |
| Production use | Officially not for production | Usable for production drills |
| Model | Workspace/Scope/Scenario/Action | Experiment/Target/Capability/Fault |
| SLA | None | Per service terms |
| Fault types | Scenario catalog (Zone Down, etc.) | Service-direct + agent-based |
| Billing | Action-minute | Action-minute |
Sources: Chaos Studio Workspaces overview and overview doc (11 November 2026). Preview has no SLA and is not for production; regions/features follow official docs.
Hidden costs: induced resources are the big line
Seven hidden costs: autoscale adding VMs/instances, database failover/replicas, extra storage/network, Azure Monitor/log ingestion, on-call/incident labour, dual-running drill environments, and the business-interruption risk of fault injection. Example: a Zone Down drill triggering 10 new VMs for 1 hour can cost more than the $3 for 30 action-minutes; database failover or cross-region traffic costs more. Chaos Studio's $0.10/minute is only the "ignition fee"; budget for the resources it ignites.
| Cost item | Trigger | Control |
|---|---|---|
| Action-minute $0.10 | Experiment actions run | Trim actions/duration |
| Autoscale | CPU/load triggers | Cap max instances/budget |
| Database failover | Failover drill | Off-peak/read replica |
| Storage/network | Cross-region/extra data | Proximity/quotas |
| Monitoring logs | Drill log spike | Sampling/retention |
| On-call labour | Drill/rollback | Scheduling/automation |
| Business interruption | Production injection | Non-prod/canary first |
Editorial summary: induced resources bill as other Azure services; this article does not estimate specific VM/DB prices; preview risk is separate.
Selection: Chaos Studio vs self-hosted Chaos Mesh/Gremlin
Chaos Studio fits teams heavily on Azure needing service-direct/agent-based faults, Azure Monitor/ARM/policy integration and no self-built control plane; action-minute billing is cheap for infrequent experiments. Self-hosted Chaos Mesh fits Kubernetes-centric teams with an existing platform team wanting full control of experiments/data plane; software is free but cluster/CRD/network/upgrades/labour are not. Gremlin and other SaaS fit cross-cloud managed fault libraries/reporting, usually quoted per seat/usage and often above action-minute pricing. Filter by fault coverage + production support + data boundary first, then unit price.
| Scenario | First choice | Check first |
|---|---|---|
| Azure resource fault injection | Chaos Studio classic | GA/region/permissions |
| Non-prod scenario trial | Workspaces preview | No SLA, not for production |
| AKS-centric | Chaos Mesh | Cluster/CRD/labour |
| Cross-cloud managed SaaS | Third party (quote) | Data/seat/fault library |
| High-frequency CI chaos | Price action-minutes first | Frequency×actions |
| Production big drill | Classic + guardrails | Induced resources/rollback |
| Tight budget | Self-hosted/small | Labour and risk |
Editorial summary, not a vendor recommendation; features/prices follow official pages; third-party/self-hosted costs are not quoted here.
FAQ
How is Chaos Studio billed?
$0.10 per action-minute in most commercial regions, based on experiment action run time; one 30-minute action is $3, three are $9. Add the induced VM/DB/storage/network/log resources.
What does one experiment cost?
Actions×duration: 3 actions×30 min = 90 action-minutes = $9; 5×60 = $30; 1×24 h = $144. Whether wait/other actions bill follows the invoice.
What do 20 experiments cost per month?
If each is 3 actions×30 min: 20×3×30=1,800 action-minutes = $180/month; if each is 5×60 min: 20×5×60=6,000 = $600/month. Induced resources are extra.
Can Workspaces be used in production?
Microsoft states Workspaces/Scenarios are public preview, with no SLA/limited warranty and not for production; use GA Experiments (classic) for production drills, or keep Workspaces in non-production.
Do parallel actions cost less?
No. Billing is per action's run minutes; three parallel 30-minute actions are still 90 action-minutes. Parallelism shortens wall-clock time, not action-minutes.
Do wait actions bill?
The official basis is action run time; this article's TCO assumes fault actions and does not separately count wait/other actions. Whether waits bill follows the actual invoice; include wait time in your budget to be safe.
What is the most overlooked cost?
Induced resources: autoscale VMs, database failover, cross-region traffic, monitoring logs and on-call labour. A drill's resource bill can dwarf a few dozen action-minutes.
How do I reduce cost?
Five steps: (1) trim actions/duration; (2) drill off-peak/non-production first; (3) cap autoscale and cross-region traffic; (4) control log sampling/retention; (5) price high-frequency CI chaos first, and consider self-hosted/low-frequency big drills.
Research method and limitations
Official facts (pricing page, Chaos Studio overview/Workspaces docs): $0.10/action-minute, billing per experiment action run time, induced resources billed separately, Workspaces/Scenarios public preview and not for production, Experiments classic GA, service-direct/agent-based faults. Editorial calculations: actions×minutes×$0.10; monthly runs×actions×duration; 1/3/5/10 actions and 5/20/50/100 runs; wait/other actions not separately counted.
Seven limitations: (1) USD list prices only, excluding tax/contract discounts/regional differences; (2) action count, duration, run frequency and whether waits bill are editorial assumptions; (3) excludes induced VM/DB/storage/network/log/labour and business-interruption risk; (4) no real fault-injection/rollback benchmark; (5) Workspaces is preview and features/regions/support may change; (6) third-party/self-hosted costs are not quoted; (7) prices/billing basis change—recheck official pages and bills before purchase.
Sources and basis
- Microsoft Azure, Azure Chaos Studio Pricing: azure.microsoft.com/en-us/pricing/details/chaos-studio. $0.10/action-minute, per experiment action run time, induced resources separate. Accessed 11 November 2026.
- Microsoft Learn, What is Azure Chaos Studio?: chaos-studio-overview. Workspaces/Scenarios preview, Experiments classic GA, service-direct/agent-based. Accessed 11 November 2026.
- Microsoft Learn, Chaos Studio Workspaces overview: chaos-studio-workspaces-overview. Workspace/Scope/Scenario/Action, preview not for production, reports. Accessed 11 November 2026.
Data cut-off 11 November 2026; prices/features follow Microsoft official pages and bills; this is not procurement, architecture or legal advice.