Tech & Energy
Fine-Tuning Cost Calculator
OpenAI charges by training tokens; inference fine-tuned models cost 50-200% more per token. Forecast the bill before you click upload.
Training cost
—
- Training tokens
- —
- Monthly inference
- —
- Break-even vs prompting
- —
FAQ
When is fine-tuning worth it? +
Three triggers: A) you have 500+ high-quality examples, B) your prompt is bloated trying to teach style, C) latency matters and a smaller model would do.
Open-source fine-tuning? +
LoRA on Llama 3.1 8B can run on a single A100 in 4-8 hours for ~$30. Self-hosted inference then costs nothing per request.