> ## Documentation Index
> Fetch the complete documentation index at: https://docs.coreweave.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Choose capacity for sandboxes

> Compare serverless capacity and CKS capacity plans for sandbox workloads, including Spot and reserved capacity.

CoreWeave Sandboxes can run on serverless capacity that CoreWeave manages or on your own CoreWeave Kubernetes Service (CKS) cluster. On CKS, sandboxes use the cluster's capacity, which can include the following options:

* [Reserved Instances](https://docs.coreweave.com/platform/capacity-plans.md#reserved-instances)
* [Flex Reservations](https://docs.coreweave.com/platform/capacity-plans.md#flex-reservations)
* [On-Demand instances](https://docs.coreweave.com/platform/capacity-plans.md#on-demand)
* [Spot Node Pools](https://docs.coreweave.com/platform/capacity-plans/spot-node-pools.md#create-a-spot-node-pool)

Spot is a capacity option for CKS sandboxes, not a separate sandbox product.

Use this page to choose capacity based on your workload's demand, interruption tolerance, and hardware requirements. After choosing, use [Sandbox placement and spillover](https://docs.coreweave.com/products/sandboxes/placement.md) to configure placement and credentials.

## Compare capacity options

| Option | When to use it | Capacity and interruption tradeoffs | Cost considerations |
| - | - | - | - |
| [Serverless](https://docs.coreweave.com/products/sandboxes/get-started.md) | Intermittent agent tool calls, evaluations, or bursts without managing cluster capacity. | CoreWeave manages placement. Access, quotas, and available resources constrain requests. Quota isn't a capacity guarantee. | For Forge software as a service (SaaS) and CoreWeave accounts, usage is billed to the account that owns the credential. For W\&B Dedicated Cloud instance keys, confirm billing terms with your account team. Compare these costs with maintaining CKS capacity. |
| [CKS with Reserved Instances](https://docs.coreweave.com/platform/capacity-plans.md#reserved-instances) | Steady sandbox demand, or suitable spare capacity you've already reserved. | Guaranteed capacity reservation during the term. The [runner](https://docs.coreweave.com/products/sandboxes/architecture.md#sandbox-resources) still requires enough free resources to place each sandbox. | A term commitment and a flat rate apply whether instances are running or idle. |
| [CKS with Flex Reservations](https://docs.coreweave.com/platform/capacity-plans.md#flex-reservations) | Variable demand that needs guaranteed peak capacity. | Reserved peak capacity. Provision enough ready capacity for incoming sandbox requests. | Holding and active-usage charges apply under your agreement. Requires a term commitment. |
| [CKS with On-Demand](https://docs.coreweave.com/platform/capacity-plans.md#on-demand) | Variable work that must avoid Spot reclamation, without a term commitment. | Not preemptible, but additional capacity might be unavailable when you need it. | Billed for actual use without a long-term commitment. |
| [CKS with Spot Node Pools](https://docs.coreweave.com/platform/capacity-plans/spot-node-pools.md#create-a-spot-node-pool) | Retryable evaluations or batch work with flexible completion times. | Capacity isn't guaranteed, and CoreWeave can reclaim it. Your application must handle interrupted work. | Reduced rates. Include retries and time spent waiting for replacement capacity in your comparison. |

All CKS options require a configured sandbox runner and a policy that permits the workload. You manage the cluster's capacity. Choosing CKS placement doesn't provision Nodes. For the commercial terms of each plan, see [About capacity plans](https://docs.coreweave.com/platform/capacity-plans.md). To run a sandbox on any of these CKS capacity options, follow [Use your own compute](https://docs.coreweave.com/products/sandboxes/get-started-cks.md).

## Check access and workload requirements

Some requirements can exclude an option before you compare demand or cost:

| Requirement | Serverless | Your CKS cluster |
| - | - | - |
| Account and credentials | A Forge API key or CoreWeave API access token. See [Choose and get credentials](https://docs.coreweave.com/products/sandboxes/placement.md#choose-and-get-credentials). | A CoreWeave API access token, sandbox permissions, and an enabled runner on your cluster. A Forge API key alone doesn't provide CKS access. |
| Resource and network policy | Policies are set at request time. | A stored runner policy governs requests. To review the policy and any deployment-configured defaults, see [Set a default policy for new runners](https://docs.coreweave.com/products/sandboxes/profiles/configure.md#set-a-default-policy-for-new-runners). |
| CPU-only isolation | CoreWeave manages the runtime. | Kata isolation applies when the configured runtime is a Kata runtime class, such as `cw-kata-qemu`. For example, set `defaultCpuRuntimeClass` in the runner policy. See [Constrain security posture](https://docs.coreweave.com/products/sandboxes/profiles/configure.md#constrain-security-posture). Check the configured runtime before running untrusted code. |
| GPU hardware | NVIDIA RTX PRO 6000 Blackwell Server Edition. | GPU types that the runner advertises from your cluster's Nodes. See [Find the GPU types on your runner](https://docs.coreweave.com/products/sandboxes/gpu-sandboxes.md#find-the-gpu-types-on-your-runner). |
| GPUs per sandbox | Up to your organization's GPUs-per-sandbox limit, which defaults to 1, subject to available capacity and the serverless policy. | Up to your organization's GPUs-per-sandbox limit, which defaults to 1, and the applicable policy's `maxGpuCount`. The request must fit on a single Node. |
| GPU isolation | GPU sandboxes run in virtual machines and support untrusted code. | GPU virtualization isn't supported for GPU sandboxes on CKS. The supported configuration uses the cluster's default runtime (`runc`), so run only trusted code. For GPU virtualization, use serverless placement or contact your account team. |

The GPUs-per-sandbox limit applies to your organization in both serverless and CKS placement. A request above it fails with `CWSANDBOX_GPU_COUNT_EXCEEDED`. To discuss a higher limit, contact support or your account team. See [GPUs per sandbox](https://docs.coreweave.com/products/sandboxes/reference/limits-and-quotas.md#gpus-per-sandbox).

GPU sandboxes require separate private-preview access for both serverless and CKS placement. For supported clients, hardware, and isolation requirements, see [Run a GPU sandbox](https://docs.coreweave.com/products/sandboxes/gpu-sandboxes.md). For CKS runtime settings, see [Configure a sandbox policy](https://docs.coreweave.com/products/sandboxes/profiles/configure.md).

The credential determines which account owns the sandbox. For Forge SaaS and CoreWeave accounts, see [CoreWeave Infrastructure and Forge billing](https://docs.coreweave.com/billing/coreweave-and-forge-billing.md). If you use a Dedicated Cloud instance key, confirm billing terms and effective limits with your account team. The Forge account defaults alone don't establish your instance's quota. See [Use sandboxes with W\&B Dedicated Cloud](https://docs.coreweave.com/products/sandboxes/dedicated-cloud.md).

Before planning a burst or fallback path, check [Choose and get credentials](https://docs.coreweave.com/products/sandboxes/placement.md#choose-and-get-credentials) and [Limits and quotas](https://docs.coreweave.com/products/sandboxes/reference/limits-and-quotas.md). CKS sandboxes don't count toward the concurrent serverless sandbox quota.

## Plan for steady demand and bursts

Separate the capacity you use consistently from occasional peaks. The following illustration shows a steady baseline with bursts above it. It represents a planning pattern, not measured usage, a price comparison, or automatic placement behavior.

![Illustrative sandbox demand over time. Demand stays near a steady baseline, with three bursts above it. A dashed line marks the baseline. The shaded area above it represents variable demand.](https://mintcdn.com/coreweave-dbfa0e8d/vrfXi6n4jsaYKjTQ/products/sandboxes/_media/capacity-demand.svg?fit=max\&auto=format\&n=vrfXi6n4jsaYKjTQ\&q=85\&s=5ae7d3ba3bbcb3c00af2acfc566eb43d)

Use these workload patterns to compare capacity options:

* **Steady demand**: Consider Reserved Instances for sustained use. If you already have suitable unused reserved capacity, evaluate using it before committing to more.
* **Variable demand with a guaranteed peak**: Consider Flex Reservations when holding peak capacity matters, even while some of it is idle.
* **Occasional bursts**: Consider serverless to avoid managing additional Nodes. On-Demand is another option when you need CKS placement and can tolerate capacity being unavailable.
* **Work that can restart**: Consider Spot when retries and flexible completion times are acceptable. Demand shape alone doesn't establish whether interruptions are safe.

You can combine CKS capacity with serverless placement. For example, use suitable CKS capacity for a baseline. Configure eligible creation requests to fall back to serverless when CKS placement fails. Both modes must support the workload. A GPU request above your organization's GPUs-per-sandbox limit fails before placement, so it doesn't fall back to the other mode.

Compare cost per completed workload, including idle capacity, retries, and operational effort. For rates and commitments, contact your account team. The demand graph alone doesn't identify the cheapest plan.

## Understand scaling and interruption limits

Node Pool autoscaling doesn't guarantee capacity for sandbox creation, and spillover doesn't recover interrupted work.

### Node Pool autoscaling

CKS [Cluster Autoscaler](https://docs.coreweave.com/products/cks/nodes/autoscaling.md) adds Nodes in response to unschedulable Pods. Sandbox placement can reject a request for insufficient capacity before creating a Pod. In that case, no pending Pod exists to trigger scale-up. A longer sandbox timeout doesn't create that trigger.

This limitation applies to sandboxes on your own CKS cluster, regardless of the Node Pool's capacity plan. Don't rely on incoming sandbox requests to scale a CKS Node Pool from zero. Provision capacity ahead of demand, keep sufficient ready capacity, or use serverless for bursts. A capacity reservation guarantees the reservation, not that Nodes are already ready to accept a sandbox request.

### Spot interruptions

Spot reclamation can interrupt a running sandbox. Use Spot for work your application can safely retry. Keep results or checkpoints needed for recovery outside the interrupted sandbox. For the Node termination sequence, see [Instance preemption](https://docs.coreweave.com/platform/capacity-plans/spot-node-pools.md#instance-preemption).

A Spot Node Pool can replenish Nodes when capacity becomes available again. Replacing a Node doesn't recover a running sandbox's state.

### Spillover between placement modes

Python's automatic spillover retries sandbox creation once in the other placement mode for eligible placement failures. It doesn't migrate running sandboxes or recover work interrupted by Spot reclamation.

Spillover requires a CoreWeave API access token for an organization with serverless access and a configured CKS runner. Each mode applies its own resource policy and limits. The TypeScript client supports both placement modes but doesn't provide automatic spillover. Creating a sandbox from a template and `run_from_file()` require strict placement without spillover.

For configuration, supported directions, and runner restrictions, see [Spill over between modes](https://docs.coreweave.com/products/sandboxes/placement.md#spill-over-between-modes).

## Configure your choice

Use the following guides to configure your chosen capacity option:

* [Get started with CoreWeave Sandboxes](https://docs.coreweave.com/products/sandboxes/get-started.md)
* [Use your own compute](https://docs.coreweave.com/products/sandboxes/get-started-cks.md)
* [Create a Spot Node Pool](https://docs.coreweave.com/platform/capacity-plans/spot-node-pools.md#create-a-spot-node-pool)
* [Request more capacity](https://docs.coreweave.com/platform/capacity-plans.md#request-more-capacity)
