Do not park your margin on the mid-tier price spread, and do not park it on "the top tier never gets cut" either, because that is no longer the standing conclusion. ENTRY 118 recorded on 7/30 that Luna fell 80% while Sol was untouched; VentureBeat reported that on 9/22 GPT-6 Sol's list price was halved to $2/$10.
The mainstream read the cut as a price war. The diary read it as a tiered price list: the bottom commoditizes faster, the top holds price and charges extra for speed, and the middle is squeezed hardest. The top-tier half of that call is now being tested by its own falsification clause, so margin has to come from your own architecture, not from an upstream tier staying expensive.
OpenAI repriced the GPT-5.6 family three weeks after launch: Luna down 80%, Terra down 20%, Sol unchanged. As reported by VentureBeat on 9/22: GPT-6 Sol at $2/$10 versus $4/$20 for the prior generation; ENTRY 118 had recorded a prior price of $5/$30 on 7/30, and the diary did not track the adjustment in between. In May Anthropic moved to API-rate billing, and effective capacity for some workflows fell 25x (community estimate, single source, not confirmed by the company).
For founders: recompute model-call cost at zero subsidy now, and if margin does not hold, change the architecture with caching, pushing work down to small models, and keeping the expensive model only at decision points. For investors: in diligence, recompute unit economics at actual API prices and watch whether the top tier follows the cuts.