GPT-6 Luna and Haiku 5.5 both list at $0.10 in, $0.50 out per million tokens. On my 20 podcast excerpts, Luna billed 40% less, and billed less on every one of the 20 requests. Gemini 3.8 Flash lists at 7.5x Haiku's price. Its bill came out 7.6x. At a 7.5x list gap, the bill tracked that gap. At a zero list gap, token counts and default thinking determined what I paid. One summary prompt, one run each. I didn't grade the answers. Run 20 of your own requests before you compare bills.