In short
Google Cloud Run: Google Cloud's managed platform for running containers, with on-demand NVIDIA GPUs attached to services.
Klio Compute: serverless GPUs for Python. Run a function on a GPU with one decorator, deploy endpoints, task queues, schedules and sandboxes, or put an open model behind an OpenAI-compatible API, and pay per second from prepaid credit.
Where Google Cloud Run is the better choice
- Any container, inside Google Cloud alongside the rest of your infrastructure.
- Billing rounded up to the nearest 100 milliseconds, and a monthly free tier for CPU and memory.
- On-demand GPUs with no reservation, and zonal redundancy as an option.
Where Klio Compute differs
- A container can have 1, 2, 4 or 8 GPUs; Cloud Run attaches one per instance.
- More GPU types: H100, A100, L40S, A10 and others as well as L4.
- Python functions with a decorator, without building or deploying a container yourself.
- Prepaid credit: work stops at zero instead of running up an invoice, and a monthly purchase limit caps what can be spent.
Features
| Feature | Klio Compute | Google Cloud Run |
|---|---|---|
| Serverless GPU functions or containers | Yes. Functions, endpoints, task queues and schedules on a GPU from one decorator. (docs) | Yes. Fully managed, on-demand GPUs. (source) |
| Scale to zero | Yes. Endpoints and models stop when idle and start on the next request. (docs) | Yes. (source) |
| Billing per second or finer | Yes. Metered per second. (docs) | Yes. Rounded up to the nearest 100 ms. (source) |
| Several GPUs in one container | Yes. 1, 2, 4 or 8 GPUs of one type; more than one needs a paid workspace. (docs) | No. One GPU per instance. (source) |
| Sandboxes | Yes. Create, run commands and code, read and write files. (docs) | Not verified. |
| Volumes | Yes. Named volumes mounted into any container. (docs) | Not verified. |
| OpenAI-compatible model serving | Yes. klio model deploy, for the models klio model list shows. (docs) | Not verified. |
| Runs Modal code with an import change | Yes. import klio_compute.compat as modal; a few features are refused with a clear message. (docs) | No. Runs your container image. (source) |
| Prepaid credit | Yes. Top up, set a monthly purchase limit; work stops at zero. (docs) | Not verified. |
GPU prices per hour
Public list prices verified 8 November 2026 from Google Cloud Run's pricing page. Klio Compute's prices are live, checked Nov 8, 08:50 UTC. GPU prices follow the market and are checked every few minutes. A running container keeps the price it started at.
| GPU | Klio Compute | Google Cloud Run |
|---|---|---|
| H100 80 GB | Not available right now | — — Cloud Run's price list has L4 and RTX PRO 6000 GPUs only. (source) |
| A100 80 GB | $2.49/h | — — Cloud Run's price list has L4 and RTX PRO 6000 GPUs only. (source) |
| L40S 48 GB | Not available right now | — — Cloud Run's price list has L4 and RTX PRO 6000 GPUs only. (source) |
| A10 24 GB | $1.95/h | — — Cloud Run's price list has L4 and RTX PRO 6000 GPUs only. (source) |
| L4 24 GB | $1.45/h | $0.67/h — Listed as NVIDIA L4, non-zonal redundancy, $0.0001867/s. (source) |
Cloud Run lists GPUs per second (L4 without zonal redundancy here); CPU and memory are billed on top.
Prices change; check Google Cloud Run's page before deciding. GPU price only, per GPU-hour; CPU, memory, storage and network may be billed separately, and discounts and committed-use prices aren't shown.
Sources
Google Cloud Run's pages read on 8 November 2026:
About this comparison
Google Cloud Run is a trademark of its owner. Klio Compute is not affiliated with it; names are used only to identify the products compared. "Not verified" means the vendor's pages we read didn't settle it either way. Corrections are welcome at support@klio.tech.