Daytona
GPU sandboxes for AI-generated code execution
Daytona provides sub-90ms sandbox spin-up for running AI-generated code, with GPU access billed per second. Pricing below is independently tracked from daytona.io/pricing.
View on Daytona →Daytona is a GPU sandbox / AI dev infrastructure provider, not a traditional dedicated GPU cloud. It's built for spinning up short-lived sandboxes to execute AI-generated code against a GPU, not for provisioning a persistent dedicated instance. Don't read the rates below as directly comparable to a dedicated-instance provider's hourly rate without accounting for that difference in what's actually being billed.
| GPU Model | Regions | On-Demand | Preemptible |
|---|---|---|---|
| B200 192GB | US | $6.25 | $3.59 |
| MI355X 288GB | US | $5.99 | $3.44 |
| H200 141GB | US, UK, EU | $4.54 | $2.61 |
| H100 80GB | US, Canada, UK, EU | $3.95 | $2.27 |
| RTX PRO 6000 96GB | US, EU | $3.03 | $1.74 |
| RTX 5090 32GB | US, Canada, Norway, EU | $1.29 | $0.74 |
| RTX 4090 24GB | US, Canada, Norway, EU | $0.99 | $0.57 |
Rates shown are the GPU compute rate only — see the billing rule below. regularly tracked · last tracked 2026-09-17 · confirm current rates before procurement.
The GPU rate above excludes vCPU, RAM, and disk — those meter separately on a per-second basis too. The on-demand and preemptible figures in the table are the GPU-only hourly rate the per-second formula is applied against, not a flat hourly charge.
Capacity is pooled and scheduled across regions — you don't pick a region by default when you spin up a sandbox. The regions listed in the pricing table above are where a given GPU is available in that pool, not a region you select per sandbox.
Pinned allocation to a specific region is available on request via [email protected].
Daytona is independently listed on GPUAdvisor. Pricing above is drawn from Daytona's own published pricing page and direct vendor confirmation — no commercial relationship influences the rates shown.