Is OpenAI Fine-Tuning free? What it actually costs in 2026
No — and the door is closing: OpenAI's pricing page says the fine-tuning platform is being wound down, is no longer accessible to new users, and existing users can create training jobs only 'for the coming months'. The one listed model, o4-mini, trains at $100 an hour.
There was never a free tier; training is billed per hour and inference for a tuned model at a premium. As of this reading the table lists a single model — o4-mini-2025-04-16 at $100 per training hour, $4 input / $16 output per million tokens ($2 / $8 with data sharing) — under a notice that new users cannot access the platform.
OpenAI's pricing page states the fine-tuning platform is being wound down and is closed to new users; one model (o4-mini) remains listed.
Any team planning to fine-tune on OpenAI must move to open-weight models; the 'when to stay' case for this target has effectively closed.
What the free tier actually includes
- Closed to new users (OpenAI's notice on the pricing page)
- Existing users: training jobs 'for the coming months'
- Tuned models stay available for inference until their base model is deprecated
Where the wall is
Access, not price. If you do not already have fine-tuning enabled, the answer to 'how much' is 'you can't'. The open route — tuning an open-weight model with Axolotl, Unsloth or LLaMA-Factory on rented GPUs — is now the only route for new work.
After that: o4-mini-2025-04-16: $100 per training hour; inference $4 input, $1 cached, $16 output per million tokens (Batch $2 / $0.50 / $8); half those inference rates with data sharing.
Can you self-host OpenAI Fine-Tuning? No — cloud only
OpenAI's models cannot be tuned outside OpenAI. Axolotl, Unsloth and LLaMA-Factory tune open-weight models on your own or rented GPUs.
What it actually takes to run one yourself
Real RAM against the documented minimum, monthly cost, setup time, and the one thing that catches people out.
When paying is still the right call
Stay when you have no ML engineer and no GPU, and the fine-tune is a means to an end rather than a product. Uploading a JSONL file is dramatically less work than standing up a training environment, and for many teams that difference is decisive. Stay also when you need the specific capabilities of a frontier base model that has no open equivalent — you cannot fine-tune your way to a model that does not exist. And stay if your volume is low enough that the inference premium is noise. The open toolchains below assume you have, or can rent, a GPU for a few hours.
What you are locked into
High, and structurally so. Your training data is portable — it is your JSONL file — but nothing else is. The adapter weights never leave OpenAI, so 'migrating' means retraining from scratch on a different base model, which is a real project rather than an export. The mitigation is to keep your training set clean, versioned and provider-agnostic from day one, because that file is the only part you actually own.
If you would rather not pay: Axolotl
Axolotl
Fine-tune most open models from one YAML file. The community default.
Free. You rent or own the GPU; a small LoRA can cost a few dollars of rented time.
What will the GPU cost? Compare every company’s hourly price
See all 5 OpenAI Fine-Tuning alternatives comparedCompare the free options head-to-head
Read from OpenAI's own pricing and documentation on 2026-09-26 — 25 days ago. Vendors change prices without notice; confirm before you commit.