Fine-Tuning Cost Calculator
Fine-tuning on a 2,000,000-token file for 3 epochs costs $30.00 on gpt-4.1-mini and $150.00 on gpt-4.1; running the tuned gpt-4.1-mini on 10 million input and 2 million output tokens a month then costs $14.40, against $7.20 for the base model (prices checked 2026-10-16).
How to use
- Pick the model to fine-tune. Only OpenAI models with a published per-token training price are listed.
- Enter the tokens in your training file and the number of epochs. The training job bills every token once per epoch.
- Enter your expected monthly inference in millions of input and output tokens. Read the training cost, the monthly bill on the fine-tuned and base models, and the table across models.
How it works
Trained tokens = training file tokens × epochs. Training cost = trained tokens × training price ÷ 1,000,000.
Monthly inference on the fine-tuned model = input millions × fine-tuned input price + output millions × fine-tuned output price.
Monthly inference on the base model uses the standard price of the same model family, so the difference is the monthly premium of running the fine-tuned model.
First-year cost = training cost + 12 × monthly fine-tuned inference. Prices are US dollars per million tokens from the OpenAI pricing page.
FAQ
Which models can I price here?
gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-4o and gpt-4o-mini, which have a training price per million tokens. o4-mini reinforcement fine-tuning is billed per hour of training, and legacy models are left out.
Why is a fine-tuned model more expensive to run?
OpenAI lists separate, higher input and output prices for fine-tuned models. On gpt-4.1-mini, for example, the fine-tuned input price is twice the base price, so savings must come from shorter prompts or a smaller model doing the job.
How many epochs should I count?
Use the number you set for the job. If you let the service choose, check the job details after it starts, because the cost scales directly with epochs.
What about Claude or Gemini fine-tuning?
The official Anthropic and Google pricing pages checked for this page list no fine-tuning prices, so those models are not included rather than estimated.