Managed fine-tuning platform
Onrup vs Together AI
Should I use Onrup or Together AI?
Together is the better choice if you want the widest model catalogue and per-token billing. Onrup is the better choice if you want a hard spend gate before compute is leased, an evaluation gate before deploy, and GPU-hour pricing you can predict.
A broad model API with fine-tuning attached, covering the widest catalogue of open-weight models of anyone in this set.
Where they differ
| Onrup | Together AI | |
|---|---|---|
| Billing model | Per GPU-second, one-minute floor | Per training token; per GPU-hour for dedicated |
| Spend control | Authorised and reserved before compute is leased | Usage limits and alerting |
| Evaluation gate | Blinded, blocking, before deploy or publish | Not part of the product |
| Model catalogue | 43 open-weight models across 13 families | Substantially larger |
| Weight portability | Download or publish to the Hub in one call | Downloadable |
On comparable hardware
Together AI publishes $5.49 per H100 80GB hour. Our serving rate for the same class is $2.99, which makes theirs 1.8× the rate. Total cost is the rate multiplied by wall time, so this ratio is the starting point of a comparison rather than the end of one.
LoRA supervised fine-tuning is $0.48 per 1M tokens for models up to 16B, rising to $1.50 for 17–69B and $2.90 for 70–100B. Full fine-tuning is roughly 2.5× the LoRA rate at every tier. Dedicated H100 capacity is $5.49 per GPU-hour on demand. Source, checked 2026-08-06.
The longer answer
These two products solve overlapping problems from opposite ends. Together starts from a very large model API and adds training to it, so the natural unit is a token and the natural experience is that infrastructure is invisible. Onrup starts from the run — a bounded job on a GPU class you chose — and the natural unit is a GPU-second.
Which is better depends almost entirely on your traffic shape. Per-token billing is excellent when usage is light or spiky, because you pay nothing for idle capacity and never think about sizing. It becomes expensive at steady volume, where the same work on a reserved GPU costs less per request than the token rate implies.
The other real difference is what happens before a run starts. Together will let a job begin and tell you what it cost afterwards. Onrup estimates the cost, reserves it against your limit, and refuses the job if it would breach the ceiling. If you have ever been surprised by a bill, that difference is the product.
On breadth, Together wins and it is not close. If the base model you need is outside the mainstream families, check their catalogue first.
Where Together AI wins
Model breadth. If the specific base model you need is unusual, Together is more likely to have it than anyone else here, and their token-priced inference means you never think about a GPU at all.
Choose them if
- You need a base model outside the mainstream open-weight families
- You would rather pay per token than reason about GPU classes
- You want fine-tuning and a large general model API from one vendor
On ownership
Weights on Onrup are downloadable from every finished run and publishable to a model hub in one call. On Together AI: Fine-tuned model weights can be downloaded.
Frequently asked questions
Is Onrup cheaper than Together AI?
On directly comparable dedicated H100 capacity, yes — Together publishes $5.49 per GPU-hour and our serving rate for the same class is $2.99. On per-token fine-tuning there is no like-for-like comparison, because we bill training by GPU-second rather than by token, and which is cheaper depends on your dataset size and sequence length.
Can I use both?
Yes, and several teams do. Both serve an OpenAI-compatible endpoint, so a client can point at either by changing a base URL. Using one for breadth of models and the other for the workloads that run constantly is a reasonable arrangement.
Do I lose model choice by moving?
Possibly. We carry 43 open-weight models across 13 families, chosen for coverage of the sizes and objectives most fine-tunes actually need. If your base model is not on that list, Together is the better fit.
Researching Together AI alternatives more broadly? →
Last verified 6 August 2026. Together 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.