Before wiring an app to a language model, know the bill: APIs charge per token, inputs and outputs at different rates, and flagship models cost 10–20× their mini versions. Enter your daily volume and see the monthly cost per model side by side.
How the math works
Monthly cost = daily calls × (avg input tokens × input rate + avg output tokens × output rate) × 30. Output tokens typically cost 3–5× input tokens, which is why verbose models quietly inflate your bill. The calculator uses publicly listed prices as approximations — check official pricing pages before committing, they change often.
Three levers that actually save money
① Downgrade the model: simple tasks (classification, extraction, rewriting) run fine on mini/flash tier at 1/10th the price. ② Shrink context: send only the necessary history — input tokens are usually half the bill. ③ Cache and batch: prompt caching charges repeated prefixes once, and batch APIs run at a discount for non-urgent jobs.