AI Fine-Tuning Cost Calculator
Estimate the one-time training cost of fine-tuning an OpenAI model on your dataset.
Training job
Training cost
$9.00
One-time cost for this training run, billed per token across all epochs.
After training, inference on your fine-tuned model is billed separately at the provider's tuned-model inference rate — check OpenAI's fine-tuning pricing page for the current rate, since it can exceed the base model's price and may change independently of training price.
How this is calculated
Token-based training: cost = dataset tokens × epochs ÷ 1,000,000 × price per million tokens. Reinforcement fine-tuning: cost = training hours × price per hour.
Pricing is verified against provider pricing pages as of 2026-09-01. AI and cloud pricing changes frequently — confirm the current rate on the provider's own pricing page before budgeting.
Frequently asked questions
Is fine-tuning training a one-time cost?
Yes — training is billed once per run, based on your dataset's token count and how many epochs you train for. Ongoing use of the fine-tuned model is billed separately, at inference time, on every request.
What's an epoch, and why does it multiply the cost?
An epoch is one full pass through your training dataset. Training for 3 epochs processes your dataset 3 times, which is why the token-based training cost scales linearly with epoch count.
Should I fine-tune, or use a bigger prompt with examples instead?
Few-shot prompting (examples in the prompt) is usually cheaper and faster to iterate on for most tasks. Fine-tuning pays off when you need consistent behavior at high volume, want to shorten prompts, or need a capability few-shot examples can't reliably teach.
Why is reinforcement fine-tuning priced differently?
Reinforcement fine-tuning (used for o-series reasoning models) trains through many sampled rollouts rather than a fixed pass over labeled examples, so it's billed by compute time (GPU-hours) instead of a flat per-token rate.