How we score fine-tuning vendors
Each vendor gets up to two points on each of five criteria, judged only from its own published pricing and documentation pages. We do not run jobs, benchmark models or audit security, so the score measures how much a buyer can verify before talking to sales.
Last verified 7 November 2026 against each vendor's own pages
What fills a square?
Full: 2 points. Partly: 1 point. Not found on the pages we read: 0. Maximum 10.
- Published pricing
Training prices with figures and units on the vendor's own page.
Price structure or plans listed, but figures not shown or not captured.
- Model choice
Several base model families or sizes listed as tunable, with methods named.
One model family only, or methods named without the models.
- Evaluation support
Evaluation built into the training workflow and documented.
Some evaluation step mentioned (for example billed validation runs, or evaluation environments).
- Data handling
Retention, training-use and deletion terms stated on the pages we read.
A single data-use term stated (for example a discount for data sharing).
- Deployment
How a tuned model is served, with its price basis published.
A serving option mentioned without its price basis.
What do we read, and what do we leave out?
For each of the 8 scored vendors we read its pricing page and, where one exists, its fine-tuning documentation. Every price on the site is copied from those pages with the date read, in the vendor's own unit.
- We do not use third-party reviews, comparison sites or analyst rankings.
- We do not fill a gap from memory. If a page did not load or was cut short, the gap is printed as such.
- A vendor that is closed to new customers is listed but not scored.
- Acquisitions are noted where they change who you would be contracting with.
Independent: no company named here owns, sponsors or has endorsed this site.
Terms changelog
Dated changes to what the scored vendors publish, newest first.
First full check of every pricing and documentation page listed on this site. Eight vendors scored.
OpenAI fine-tuning marked as winding down and shut to new sign-ups on its API pricing page. Taken off the scored list.
CoreWeave announced it was acquiring OpenPipe.
Prime Intellect launched its Environments Hub for RL and evaluation environments.
Questions about the method
Can a vendor pay for a better score?
No. A square is filled only by what the vendor's own page states. Introduction fees, where a vendor pays them, are not a criterion and are never discussed with vendors as one.
Why not test the platforms?
A fair benchmark needs the same data, model and evaluation on every platform, and results would only hold for that one task. What a buyer can always check is what a vendor commits to in writing, so that is what we score.
What if a vendor publishes something we missed?
The next check picks it up and the change is dated in the changelog below. An empty square means "not found on the pages we read", never "does not exist".
Why are some prices left out?
Where a page loaded its figures in the browser (Azure OpenAI) or was only partly captured (AWS Bedrock), we print only what we recorded and say so, rather than fill gaps from memory.
Get a quote
Start with a ZIP code, add an email, then tap through what you want tuned. We read it, check which vendors' published terms fit, and write back with names before anyone else sees your brief.
This is an introduction service. The vendor you are introduced to may pay it a fee if you sign work together; the team asking is never charged.
Brief received.
- We read it against the published terms, model lists and deployment options of the vendors on our scorecard.
- We write back with the vendors that fit and why, by name. Nothing has gone to any of them yet.
- Want an introduction? Tell us which vendor. Your brief and contact details move only after that reply.