Two models list the same price: $0.10 in, $0.50 out per 1M tokens. I ran 20 podcast excerpts through both. GPT-6 Luna cost 40% less than Haiku 5.5 on this job, cheaper on 20 of 20 requests. On code generation their gap was 12x, on my summarization job 40%. The job determines the size. For me it came from token counts for the same text, plus default thinking charged as output. For the same text, Haiku counts 119,723 input tokens and Luna 87,055. Haiku charged for 2.1x Luna's output tokens,…