Home / GPU sandboxes / Thunder Compute / Pricing
Thunder Compute pricing
In short (2026-11-13): $0.051 per vCPU-hour, $0.016 per GiB-hour of memory. Free tier: No free tier or trial credit is stated on the pricing or billing pages. Source: official pricing page. Prices checked 2026-11-15.
Usage rates
- vCPU
- $0.051 per vCPU-hour
- Memory
- $0.016 per GiB-hour
- Storage
- $0.0876 per GiB-month
- Billing
- per second (sandboxes); per minute (instances). Sandboxes: each resource per second until the sandbox stops or its TTL ends. Instances: while running; snapshots bill until deleted.
GPU prices
- H100 SXM (sandbox): $2.95 per hour
- A100 80GB (sandbox): $1.95 per hour
- H100 SXM (instance, 4 vCPU included): $3.20 per hour
- A100 80GB (instance, 8 vCPU included): $1.09 per hour
- L40 (instance, 6 vCPU included): $0.79 per hour
- RTX A6000 (instance, 6 vCPU included): $0.35 per hour
What it costs in practice
| Workload | Usage | Cheapest plan | Total per month |
|---|---|---|---|
| Hobby agent: 1 vCPU, 1 GiB, 20 hours a month | $1.34 | Pay as you go | $1.34 |
| One dev agent: 2 vCPU, 4 GiB, 8 h a day for 22 days | $29.22 | Pay as you go | $29.22 |
| Team: 10 sandboxes of 2 vCPU, 4 GiB, 8 h a day for 22 days | $292.16 | Pay as you go | $292.16 |
| Always on: 4 vCPU, 8 GiB, 730 hours a month | $242.36 | Pay as you go | $242.36 |
CPU busy 30% (50% always on), which matters only where CPU is billed by active time. Storage, network and GPU are not included. Try your own numbers in the cost calculator.
Billing details that change the bill
- The vCPU, memory and storage rates shown are sandbox rates; storage is $0.00012 per GiB-hour, shown per month as x 730 hours. CPU-only sandboxes pay only these rates.
- Sandboxes are not open to everyone yet: access must be enabled for your organization after a request. Instances are self-serve from the console.
- Instance prices are per GPU-hour and include a set number of vCPUs and RAM; extra vCPUs cost $0.04 per vCPU-hour. The first 100 GB of instance disk per GPU is included, then $0.219 per GB-month while running; instance snapshots cost $0.05 per GB-month.
- There are no data egress fees. Instances and sandboxes are never preempted.
- Sandbox disks are wiped when the sandbox ends; instances keep their disk and support snapshots.
- Sandbox outbound internet can be left open, closed, or limited to domain and CIDR allowlists, and changed on a running sandbox.
- Maximum sandbox vCPU and memory are set by a resource catalog that is not published on the pages checked; instances allow up to 16 vCPUs per GPU on A100 and H100.
Price history
H100 per hour over timeNo change since 2026-11-13
No price change found since we started tracking on 2026-11-13. We check every pricing page daily.
Cheaper or better-fit Thunder Compute alternatives →
Sources
- https://www.thundercompute.com/pricing, checked 2026-11-13:
|vCPU |$0.051 /vCPU-hr $0.0000142 /vCPU-sec |
- https://www.thundercompute.com/pricing, checked 2026-11-13:
|Memory |$0.016 /GiB-hr $0.00000444 /GiB-sec |
- https://www.thundercompute.com/pricing, checked 2026-11-13:
|Storage |$0.00012 /GiB-hr $0.0000000333 /GiB-sec |
- https://www.thundercompute.com/pricing, checked 2026-11-13:
|H100 SXM |$2.95 /GPU-hr $0.000819 /GPU-sec |
- https://www.thundercompute.com/pricing, checked 2026-11-13:
|A100 80GB |$1.95 /GPU-hr $0.000542 /GPU-sec |
- https://www.thundercompute.com/pricing, checked 2026-11-13:
|H100 SXM |80 GB |4 included · 16 max |
- https://www.thundercompute.com/pricing, checked 2026-11-13:
$0.35 /hr
- https://www.thundercompute.com/pricing, checked 2026-11-13:
Request access and our sales team will follow up to get you set up.
- https://www.thundercompute.com/pricing, checked 2026-11-13:
No, Thunder Compute does not bill for data egress.
- https://www.thundercompute.com/docs/sandboxes/quickstart, checked 2026-11-13:
Thunder sandboxes are isolated Ubuntu microVMs powered by Firecracker, not gVisor containers.
- https://www.thundercompute.com/docs/sandboxes/quickstart, checked 2026-11-13:
Sandbox access must be enabled for your organization.
- https://www.thundercompute.com/docs/faq/faq, checked 2026-11-13:
Instead of reserving a GPU, cloud instances on Thunder share a pool of network-attached GPUs.
- https://www.thundercompute.com/products/sandboxes, checked 2026-11-13:
Close the network for evals so the model can’t look answers up.