Compute
Serverless GPU
Run containers on GPUs that scale to zero.
What it is
Serverless GPU platforms run your code or container on a GPU only while a request is being processed, scaling from zero to many replicas automatically. You pay per second of GPU time, with no idle servers.
How pricing works
Per GPU-second (we show the hourly equivalent) plus CPU/memory. Cold starts take time but are usually not billed at full rate.
We compare everything in one unit: USD per GPU-hour (billed per second).
Typical uses
- Spiky inference workloads
- Custom model APIs
- Background jobs triggered by users
What to compare
- Cold-start time
- Max concurrency
- Supported GPUs
- Developer experience
Watch out for
- Hourly-equivalent prices are higher than VMs; serverless wins only when utilisation is low or spiky.
All 5 APIs, cheapest first
Free to browseTick 2 to 4 APIs to compare them side by side.
| API | $ / GPU-hr | GPU | VRAM | Typical cold start | Quality | Price data | Compare |
|---|---|---|---|---|---|---|---|
| Baseten T4 Baseten | $0.63 | T4 | 16 GB | — | – | Verified Oct 6, 2026 | |
| Modal A10G Modal | $1.10 | A10G | 24 GB | 4 s | 65 | Unverified sample | |
| RunPod Serverless A100 RunPod | $2.72 | A100 80GB | 80 GB | 8 s | 80 | Verified Oct 4, 2026 | |
| Modal H100 Modal | $3.95 | H100 SXM | 80 GB | 5 s | 88 | Unverified sample | |
| Baseten B200 Baseten | $9.98 | B200 | 180 GB | — | – | Verified Oct 6, 2026 |