Pricing and budgets

OpenPond separates the subscription plan from metered sandbox resources. Your plan controls included usage, concurrency, retention limits, and product access. A running sandbox accrues compute and VM-disk usage from its start time until it is stopped or deleted. Durable volumes accrue storage usage while retained.

The authenticated Billing page and the SDK pricing endpoint are authoritative for the rate card applied to your account. The current public retail rates are:

Resource	Meter	Current rate	Hourly equivalent
vCPU	vCPU-second	$0.000020	$0.072000 per vCPU
Memory	GiB-second	$0.0000065	$0.023400 per GiB
VM disk	GiB-second	$0.0000000333	$0.000120 per GiB
Durable volume	GiB-second while retained	$0.00000002	$0.000072 per GiB

Model inference and connected services have their own usage rules. A sandbox budget limits sandbox execution; it is not a promise that every other service used by a workflow is included in that amount.

Resource tier examples

These estimates assume the sandbox remains running continuously. Real ephemeral Work should normally cost much less because it deletes compute after each completed turn.

Tier	vCPU	Memory	VM disk	Estimated hourly cost	30-day continuous estimate
Tiny	0.25	0.5 GiB	4 GiB	$0.030180	$21.729600
Small	0.5	1 GiB	8 GiB	$0.060360	$43.459200
Default	1	2 GiB	10 GiB	$0.120000	$86.400000
Builder	2	4 GiB	20 GiB	$0.240000	$172.800000
Heavy Builder	4	8 GiB	40 GiB	$0.480000	$345.600000

Example: a Builder sandbox running for eight minutes costs approximately $0.240000 × 8 / 60 = $0.032000. Billing uses per-second quantities and produces a receipt, so the final charged amount comes from measured duration rather than this rounded example.

Plans and included usage

Plan	Base price	Included hosted usage	Concurrent sandboxes	Retained sandboxes
Free	$0	Limited monthly usage	1	5
Pro	$20/month	$20 monthly usage	10	100
Enterprise	Custom	Contract-defined	Contract-defined	Contract-defined

Included usage is applied before the prepaid balance and opt-in metered overage. Account spend caps and overage policy can stop additional allocation even when a request-level budget is higher.

Use every budget layer

For raw sandboxes and runtimes, set both the reservation budget and execution quotas:

const sandbox = await openpond.sandboxes.create({
  resources: { cpu: 1, memoryGb: 2, diskGb: 10 },
  budget: { maxUsd: "0.20" },
  quotas: {
    maxSpendUsd: "0.20",
    maxDurationSeconds: 600,
    idleTimeoutSeconds: 180,
    maxCommands: 20,
    maxOpenPorts: 1,
    maxSnapshots: 0,
  },
});

The controls answer different questions:

budget.maxUsd bounds the reservation.
quotas.maxSpendUsd bounds execution spend.
maxDurationSeconds limits total running time.
idleTimeoutSeconds is crash protection, not successful-turn cleanup.
Command, port, and snapshot quotas constrain expensive or externally visible operations.

Stop, delete, and storage cost

Stopping ends active compute billing but intentionally retains recoverable sandbox state. Deleting releases the sandbox and should be the default after exported one-off work. Snapshots, archives, and durable volumes can continue consuming retained-storage quota or storage charges after compute stops.

For Work, use cleanup: "delete" with persistOutput. For raw sandboxes, delete in finally. For runtimes, release the current sandbox after the active step and archive the runtime when the workflow itself is complete.

Continue with workflow modes or runtimes, snapshots, and cleanup.