Max Woolf compares recent LLM price cuts and analyzes competition, inference efficiency and cache pricing. Rather than a uniform reduction, Opus 5.5 lowers input and output rates by 20% and cached-input rates by 60%. Coding workflows that repeatedly reuse context can be particularly sensitive to cache prices, but not every request becomes cheaper by the same proportion. Interpretations of competitors’ motives are the author’s analysis; actual cost depends on model choice, cache hits and usage volume.