Which models can be fine-tuned with OpenAI?

The common picks for fine-tuning are GPT-4o-mini and GPT-3.5. Vision fine-tuning is also supported, broadening use cases beyond pure text.

What are the main trade-offs of using OpenAI Fine-tuning?

It is pricier than open-model fine-tuning and does not allow weights export, meaning you are locked into OpenAI hosting. Only certain base models are eligible for fine-tuning.

How does the fine-tuning process work?

Users upload a JSONL file of example conversations and kick off a training job. The resulting custom model is hosted on the same infrastructure as the base model with the same SLAs.

Is OpenAI Fine-tuning suitable for teams needing multi-cloud portability?

No, it is not recommended if you need weights export or multi-cloud portability. It is best for teams already on the OpenAI API who want a seamless integration.

OpenAI Fine-tuning

✓ Editorially verified

Fine-tune GPT-4o-mini and friends on your own data.

Paid· Basic: $10 · Pro: $25 · Enterprise: Contact salesFine-tuningGPT-4o-mini / GPT-3.58.4 / 10

Visit website →

In short

OpenAI Fine-tuning enables users to train custom models on specific data, baking in style, format, or domain knowledge. It is best for teams already on the OpenAI API seeking the simplest path to a custom model without managing infrastructure.

Best for

Pick OpenAI FT when you're already on OpenAI's API and want the simplest path to a custom model.

Skip if

Skip it if you need weights export, multi-cloud portability, or aggressive cost control.

OpenAI's fine-tuning offering is the easiest way to fine-tune a closed frontier model. Upload a JSONL of example conversations, kick off a training job, get a custom model that bakes your style, format, or domain knowledge into the weights. Vision fine-tuning is now supported, broadening the use cases beyond pure text.

The ergonomics are excellent. The fine-tuning UI walks you through dataset prep; the API supports automation; the resulting custom model is hosted on the same infrastructure as the base model with the same SLAs.

The trade-offs are well-known: pricier than open-model FT, no weights export (you're locked into the OpenAI hosting), and only certain base models are FT-eligible (GPT-4o-mini and GPT-3.5 are the common picks). For teams already on OpenAI's API, the integration is seamless.

Editor's take

OpenAI's fine-tuning is the path of least resistance for teams already on the platform. The vision FT addition broadened the use cases; the lock-in remains the same trade-off it's always been.

— The AI Tool Bible editorial team

Pros

✅ Easiest fine-tuning UX
✅ Vision FT now supported
✅ Works inside the OpenAI ecosystem
✅ Same infra/SLA as base models

Cons

⚠️ Pricier than open-model FT
⚠️ No weights export

Use cases

styleformatdomain knowledge

Frequently asked

Which models can be fine-tuned with OpenAI?: The common picks for fine-tuning are GPT-4o-mini and GPT-3.5. Vision fine-tuning is also supported, broadening use cases beyond pure text.
What are the main trade-offs of using OpenAI Fine-tuning?: It is pricier than open-model fine-tuning and does not allow weights export, meaning you are locked into OpenAI hosting. Only certain base models are eligible for fine-tuning.
How does the fine-tuning process work?: Users upload a JSONL file of example conversations and kick off a training job. The resulting custom model is hosted on the same infrastructure as the base model with the same SLAs.
Is OpenAI Fine-tuning suitable for teams needing multi-cloud portability?: No, it is not recommended if you need weights export or multi-cloud portability. It is best for teams already on the OpenAI API who want a seamless integration.