Hyperscaler ML platform
Onrup vs AWS SageMaker
Should I use Onrup or AWS SageMaker?
SageMaker is the better choice when procurement, data residency or existing commitments confine you to AWS. Onrup is the better choice when they do not, because the fine-tuning path is shorter and the cost is legible.
The full machine-learning platform on AWS. Enormously capable, correspondingly heavy, and priced per instance-hour.
Where they differ
| Onrup | AWS SageMaker | |
|---|---|---|
| Scope | Fine-tuning and inference | The full machine-learning platform |
| Time to first run | Upload a dataset, pick a template | Roles, buckets, images, instance families |
| Pricing | Published rate card per GPU class | Per instance-hour, by family and region |
| Data residency | Tenant-scoped object storage | Your own account and region |
| Evaluation gate | Blinded, blocking, before deploy | Build it yourself |
On price
Billed per instance-hour by instance family and region, with H100 capacity sold as multi-GPU instances rather than single cards, so no per-GPU-hour figure is directly comparable.
Rather than estimate a comparable figure, we leave it blank. Their pricing page, checked 2026-08-06.
The longer answer
SageMaker can do everything, and that is both the recommendation and the warning. Almost any machine-learning workflow can be expressed in it. The cost is that expressing a simple one requires engaging with roles, buckets, container images, instance families and a pricing model with several hundred line items.
For a team whose job is one fine-tune, that overhead dominates. For an organisation running a portfolio of models with governance requirements attached, it is the point.
The decisive factor is usually not technical. If your data cannot leave your AWS account, or procurement has already approved AWS and would take two quarters to approve anyone else, the comparison is over before it starts and we would say so.
Where the constraint does not bind, the difference is time to first run and the ability to answer "what did that cost" without a report.
Where AWS SageMaker wins
It is already approved. If your organisation has an AWS agreement, a security review that took nine months and data that is not allowed to leave the account, none of the rest of this comparison matters.
Choose them if
- Data residency or procurement rules confine you to AWS
- You have committed spend to burn down
- You need the surrounding platform, not just fine-tuning
On ownership
Weights on Onrup are downloadable from every finished run and publishable to a model hub in one call. On AWS SageMaker: Artefacts land in your own S3 bucket.
Frequently asked questions
Can I keep my data in my own AWS account?
Not today. Datasets live in tenant-scoped storage on our side. If data residency in your own account is a hard requirement, SageMaker is the correct answer.
How does the cost compare?
Not directly comparable — they price per instance-hour by family and region, often with multiple GPUs per instance, so there is no single number to hold against our per-GPU-hour rate. Ours is simpler to forecast; whether it is cheaper depends on the instance you would have chosen.
Do you integrate with SageMaker?
No direct integration. Weights are downloadable, so deploying a model trained here into your own AWS environment is straightforward.
Researching AWS SageMaker alternatives more broadly? →
Last verified 6 August 2026. AWS SageMaker figures come from their own pricing page on the date checked.
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.