Rent a GPU by the hour, or hand us a model and we serve it behind an OpenAI-compatible API. Instances from $0.19 an hour. No contracts, no minimum spend, billing per second.
2.1M GPU-hours served since 2023 · no credit card required to start
Rent a raw GPU, host a model, or both. Either way it is tracked in one console and billed per second.
$0.19 /hour
from, RTX 4090
Root access, persistent disk, public IPYour image or one of oursUp and online in a few minutesStop billing the moment you delete
$29 /month
plus per-token usage
OpenAI-compatible endpointYou pick the GPU, we run itAutoscaling with per-second compute billingMost Llama, Qwen, Mistral checkpoints deploy fastThree steps, none of them involve talking to a salesperson.
4090 for a few bucks a day, H100 for real training runs. Create it, it is booted within minutes, you get an IP.
SSH in and run your container, or upload a model and our hosting serves it for you. Either way, your code stays yours.
Billing stops the moment the instance is gone. A weekend batch job costs a few dollars, and you keep your storage.
Evaluation labs, product companies serving a model to their app, researchers who need a GPU for a weekend. A few bigger teams run batch jobs with us because per-second billing makes idle time nearly free.
"We moved a 7B to their hosted API and the bill is smaller than the Kubernetes cluster it replaced. The spending cap is what actually sold us."
Marco Villanueva, ML lead, OCTO Technology
"Rented an H100 on a Friday, fine-tuned over the weekend, paid about $150, deleted it. Nobody else makes that this easy."
Priya Raman, research engineer, Seldon
GPU-hours served since 2023
models hosted on our inference service
regions: Frankfurt, Ashburn, Singapore
people on the team, and someone is always on call
uptime over the last 90 days
cheapest hour of compute you can rent
No credit card, no contract, no call. Create an account and the credit is there.
Create your account