Vast.ai rental costs: look beyond the hourly GPU price.

1. Estimate the work, not just the GPU

For a hypothetical offer priced at $0.25 per hour, an eight-hour run is $2 before costs outside that hourly rate. Twelve such runs would be $24. These are example inputs, not a current Vast.ai quote.

If the displayed price already covers all GPUs in the rental, do not multiply it by GPU count again. Two GPUs running eight hours produce 16 GPU-hours, but the price calculation still uses the offer’s total hourly rate.

Allow time for your setup, model downloads, retries, and saving results. Compare offers against the same workload and required memory, not just the cheapest headline number.

Try those inputs in the free rental cost calculator →

2. Check what the displayed rate includes

Vast.ai lists separate compute, storage, and bandwidth costs. Rates vary by machine; inspect the offer’s pricing breakdown before approving it. Only add a charge separately if your chosen hourly figure does not already include it.

Keep a short checklist: total hourly rate, intended runtime, disk allocation, expected data transfers, and a retry allowance. Treat the result as an estimate—not a provider billing limit.

Source: Vast.ai billing documentation.

3. Stop is not the same as destroy

Stopping a Vast.ai instance preserves its container data, but storage charges continue. Destroying the instance deletes its container data permanently. Copy your outputs somewhere safe and verify the backup before destroying anything.

Separate persistent volumes can survive instance destruction and have independent billing. Review those resources too; removing one instance does not necessarily remove every billable resource in your account.

Sources: Vast.ai instance management and storage types.

4. Verify cleanup in Vast.ai

After a job, check the provider console for remaining instances and storage. A local app closing, a disconnected session, or a finished workload is not evidence that the provider released your resources.

Keep the instance ID with your job notes so you can match the rental to its outputs and billing. Check the provider’s current documentation when behavior or pricing is unclear.