Cheap LLM APIs: align the billing unit first, then compare unit price

Two gateways can quote 1/10 of list price and still differ 10x on your real bill. Compare the billing unit, the included tokens and the failure policy before you compare the price.

• Three billing models exist side by side: per token, per call, per second of output.
• Four dimensions that actually decide cost: unit, cache, retries billed or not, minimum charge.
• How to verify yourself: one prompt, two providers, same output length, compare the invoice.