Cost per task, not price per million tokens, decides your bill, because reasoning models can spend 3.11x to 14.78x more tokens.
Since August, DeepSeek moved to peak and off-peak rates, Anthropic cancelled the Sonnet 5 price rise, and Claude Haiku 5.5 at $0.10 and $0.50 undercuts DeepSeek Flash list prices.
Prompt caching is the largest lever: 50x off on DeepSeek Flash, 20x on Sonnet 5.5, 10x on Haiku 5.5.