Closed model API
Onrup vs OpenAI fine-tuning
Should I use Onrup or OpenAI fine-tuning?
OpenAI fine-tuning is the better choice if frontier base quality is the requirement and portability is not. Onrup is the better choice if you need to own the weights, control the cost, or run the model somewhere OpenAI does not.
Fine-tuning of OpenAI’s own closed models, served only from OpenAI. The default first stop, and the thing most teams are trying to leave.
Where they differ
| Onrup | OpenAI fine-tuning | |
|---|---|---|
| Weight ownership | Download or publish at any time | Never exportable |
| Base models | 43 open-weight models | OpenAI's own closed models |
| Where it runs | Your endpoint, or anywhere you take the weights | OpenAI's infrastructure only |
| Billing | Per GPU-second against an authorised limit | Per training and inference token |
| Evaluation gate | Blinded, blocking, before deploy | Evaluations available, not blocking |
On price
Priced per training token and per inference token by model, with fine-tuned inference charged above the base rate. Rates are not reproduced here because the pricing page could not be retrieved at the date checked; see the source for current figures.
Rather than estimate a comparable figure, we leave it blank. Their pricing page, checked 2026-08-06.
The longer answer
Most people arriving at an open-weight platform have come from here, and usually for one of three reasons: the inference bill, a portability requirement that turned up in a contract, or a deployment environment that OpenAI cannot reach.
It is worth being clear about what you give up. OpenAI’s base models are strong and the fine-tuning API is about two calls. Nobody has to think about VRAM, GPU classes or sequence lengths. For a team without machine-learning capacity and without a portability requirement, that is a genuine advantage and the honest recommendation is often to stay.
What you gain is ownership and predictability. The weights are a file you hold. The model runs on an endpoint you control, or on hardware in your own building if you take the weights there. And the cost is a GPU-hour rate rather than a token rate, which is materially cheaper at volume and much easier to forecast.
The usual pattern that works is narrowing. A frontier model doing one job on one distribution can generally be replaced by a fine-tuned 8B with no measurable quality loss. A frontier model doing everything cannot, and no amount of fine-tuning changes that.
Where OpenAI fine-tuning wins
Quality per unit of effort. The base models are strong, the fine-tuning API is two calls, and nobody has to think about VRAM. For a team without ML experience and without a portability requirement, that is a real advantage.
Choose them if
- The frontier model quality is the requirement and nothing else is
- You have no ML capacity and want the shortest possible path
- Model portability genuinely does not matter to your business
On ownership
Weights on Onrup are downloadable from every finished run and publishable to a model hub in one call. On OpenAI fine-tuning: The weights are never yours and cannot be exported. That is the structural difference between the two products, and it is worth deciding on before you build a dataset rather than after.
Frequently asked questions
Will an open-weight fine-tune match GPT-class quality?
On a narrow, well-defined task with good training data, frequently yes — and the evaluation gate is how you find out before committing rather than after. On general open-ended capability, no, and anyone telling you otherwise is selling something.
How much cheaper is it really?
It depends on your volume and prompt length, and the honest answer is that it varies too much to quote a multiple. The structural difference is that a long system prompt repeated on every request disappears into the weights, and at high volume that prompt is usually where the money was going.
Can I migrate my existing fine-tune?
Not the model — those weights cannot be exported. You migrate the training data, which you still have, and retrain against an open-weight base.
Researching OpenAI fine-tuning alternatives more broadly? →
Last verified 6 August 2026. OpenAI fine-tuning 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.