Fine-Tuning Calculator

This calculator tells you if fine-tuning or prompting is more expensive.

Let's assume you were using GPT 4 for basic prompting, and with fine-tuning, you're now using GPT 3.5 Turbo.

”If you can achieve good results with GPT-4, you can often reach similar quality with a fine-tuned gpt-3.5-turbo model by fine-tuning on the GPT-4 completions, possibly with a shortened instruction prompt.”

- OpenAI Docs

According to the OpenAI Pricing Page, the cost of regular prompting is:

$0.0015 / 1K input tokens

$0.002 / 1K output tokens

The cost of fine-tuning is:

$0.008 / 1K training tokens

$0.012 / 1K input tokens

$0.016 / 1K output tokens

Run Count

100,000

(1 — 20,000,000)

System Prompt Tokens

500

(1 — 4,000)

Average User Prompt Tokens

500

(1 — 4,000)

Fine Tuning Examples

(1 — 1,000)

Average Output Tokens

(1 — 4,000)

Savings using Fine Tuning:

$2,487.67

Basic Prompting

Input Cost:

$3,000.00

Output Cost:

$120.00

Cost:

$3,120.00

Fine Tuning

Training Cost:

$0.33

Input Cost:

$600.00

Output Cost:

$32.00

Cost:

$632.33

Formula

BASIC_INPUT_COST = 0.03 / 1000

BASIC_OUTPUT_COST = 0.06 / 1000

FT_TRAINING_COST = 0.008 / 1000

FT_INPUT_COST = 0.012 / 1000

FT_OUTPUT_COST = 0.016 / 1000

NUM_EPOCHS = 4 (default number of epochs)

basicInputCost = (systemPromptTokenCount + userPromptTokenCount) * BASIC_INPUT_COST * runCount

basicOutputCost = outputTokenCount * BASIC_OUTPUT_COST * runCount

basicTotalCost = basicInputCost + basicOutputCost;

fineTuningTrainingCost = (systemPromptTokenCount + userPromptTokenCount + outputTokenCount) * FT_TRAINING_COST * examplesInFineTuningCount * NUM_EPOCHS

fineTuningInputCost = userPromptTokenCount * FT_INPUT_COST * runCount

fineTuningOutputCost = outputTokenCount * FT_OUTPUT_COST * runCount

fineTuningTotalCost = fineTuningTrainingCost + fineTuningInputCost + fineTuningOutputCost;