What generating synthetic training rows with an LLM costs.
Bulk generation is cheap; the tokens lost to rejected and duplicate rows are the real spend. A tight prompt and a cheap generator win. Size the calls on the LLM token cost calculator and compare against fine-tuning.
Free synthetic data cost calculator — rows, tokens per row and model price to the cost of generating a dataset.
Rows times tokens per row times the model rate, plus the tokens burned on rejected generations you filter out. A cheap model at a dollar per million tokens makes tens of thousands of rows for a few dollars — the reject/retry rate is what moves the number.
Use a small, cheap model for bulk generation and reserve a stronger model for validation, tighten the prompt so fewer rows are rejected, and de-duplicate. The waste is in retries and near-duplicate rows, not the base generation.