Hyperscaler ML platform
Onrup vs Google Vertex AI
Should I use Onrup or Google Vertex AI?
Vertex AI is the better choice if your data is already in Google Cloud or you want to tune Gemini specifically. Onrup is the better choice for open-weight fine-tuning where the weights have to be portable.
Google Cloud’s machine-learning platform, with tuning for Gemini and for a set of open-weight models.
Where they differ
| Onrup | Google Vertex AI | |
|---|---|---|
| Base models | 43 open-weight models | Gemini plus a set of open-weight models |
| Weight portability | Always downloadable | Open-weight tunes export; Gemini tunes do not |
| Data gravity | Tenant-scoped object storage | Native to BigQuery and Cloud Storage |
| Pricing | Published rate card per GPU class | Per node-hour or per token, by region |
| Evaluation gate | Blinded, blocking, before deploy | Evaluation services, not blocking |
On price
Priced per node-hour and per token depending on the product path, varying by region; there is no single comparable H100 hourly rate.
Rather than estimate a comparable figure, we leave it blank. Their pricing page, checked 2026-08-06.
The longer answer
Vertex’s strongest argument is distance. If your data is in BigQuery, the path from a table to a training job is shorter there than anywhere else, and that distance is where a surprising number of projects actually die — not in modelling, in data movement.
The portability picture is split and worth understanding before you start. Tuning an open-weight model on Vertex produces artefacts you can export. Tuning Gemini does not, and that is a decision you make at the beginning that becomes expensive to revisit.
Our argument is narrower: a curated catalogue where every model is exportable, a rate card you can read in one screen, and a gate between training and deployment. Vertex is a much larger product with a much larger surface.
Where Google Vertex AI wins
If your data is already in BigQuery, the distance from data to trained model is shorter here than anywhere else, and that distance is usually where projects actually die.
Choose them if
- Your data already lives in Google Cloud
- You want to tune Gemini rather than an open-weight model
- Procurement confines you to GCP
On ownership
Weights on Onrup are downloadable from every finished run and publishable to a model hub in one call. On Google Vertex AI: Open-weight tunes export; Gemini tunes do not.
Frequently asked questions
Can I tune Gemini here?
No. The catalogue is open-weight models only, which is the same reason everything trained here is downloadable.
What if my data is in BigQuery?
Export the training set as JSONL and upload it. For a one-off fine-tune that is a small job; for a continuously retrained model, the data gravity argument favours staying on Vertex.
How do the costs compare?
Vertex prices per node-hour or per token depending on the path, varying by region, so a like-for-like number does not exist. Ours is a flat published rate per GPU class with no regional variation.
Researching Google Vertex AI alternatives more broadly? →
Last verified 6 August 2026. Google Vertex AI 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.