What can you use now that OpenAI fine-tuning is closing?

As of November 2026, OpenAI’s own API pricing page says “OpenAI is winding down the fine-tuning platform”, and new users can no longer sign up for fine-tuning. If you need OpenAI models, Azure OpenAI still lists tuning for the GPT-4.1 and GPT-4o families and o4-mini. Otherwise, the alternative is to tune an open model on Together AI, Fireworks AI, Google Vertex AI or AWS Bedrock.

What exactly is changing at OpenAI?

According to OpenAI’s API pricing page, fine-tuning is closed to new users. Teams already on it may keep creating training jobs “for the coming months”, and their tuned models stay callable until each base model is deprecated. For a project starting now, OpenAI is not an option, which is why our scorecard lists it without a score.

If you already have tuned OpenAI models in production, plan the move now. A tuned model depends on its base model, so when the base model goes, the tuned one goes with it.

The alternatives, on published terms

If you need	Look at	What its pages publish
OpenAI models specifically	Azure OpenAI	Tuning for GPT-4.1, 4.1-mini, 4.1-nano, GPT-4o, 4o-mini, and reinforcement tuning for o4-mini; hourly hosting per tuned model; figures did not load for us
The widest open-model choice	Together AI	SFT and DPO prices per model, from $0.34 per 1M tokens (Llama 3.1 8B); minimum $4 to $60 per job
A simple price sheet	Fireworks AI	LoRA SFT from $0.50 to $10.00 per 1M training tokens by model size; no serving surcharge
Gemini or Google Cloud	Google Vertex AI	Gemini tuning per 1,000 tokens; open models such as Llama 3.1 8B at $0.67 per 1M
AWS, or reinforcement tuning	AWS Bedrock	RFT on gpt-oss-20b and Qwen3 32B at $80.00 per training hour, served per token

Vendors on this list can owe this site a fee if an introduction it makes ends in a signed contract. Rows, order and wording are set without regard to that. Independent: no company named here owns, sponsors or has endorsed this site.

None of these is a drop-in replacement. Moving from a tuned OpenAI model to an open model means retraining on your data and measuring again.

How to move a fine-tuned OpenAI project

1. Export what you own. Keep your training files and, above all, your evaluation set. They are what you will retrain and judge with.
2. Record the current baseline. Score your tuned OpenAI model on the evaluation set while it still runs.
3. Shortlist two base models. One about the size you think you need, one smaller.
4. Retrain with LoRA first. It is the cheapest test, and Fireworks AI prices LoRA at half its full-tuning rate.
5. Compare on cost per call as well as score. Serving decides the monthly bill; price it at your real traffic.

Does xAI’s Grok API support fine-tuning?

We have no published xAI fine-tuning terms in our research, so we cannot say. Check xAI’s own API documentation, and treat any answer that is not on its pages as unconfirmed.

Best fine-tuning company for a small business

For a small team, the cheapest path is usually a platform with low per-job minimums and per-token prices, plus a few hundred well-reviewed examples. At Fireworks AI’s LoRA rate for models up to 16B, a run of a few million training tokens costs a few dollars. What costs real money is preparing the data and building the test, and that is where a small business may want outside help. Our list of best LLM fine-tuning companies separates platforms from firms that do that work.

Questions to ask any replacement

• Can we export the tuned weights or adapter if we leave?
• How is the tuned model served, and does it bill while idle?
• Is our training data kept after the job, and is it used for anything else?
• What is the deprecation policy for the base model we choose?

The questions-to-ask post covers the rest. Choose the replacement you could leave again.