GPT-6 Luna First, Sol on Demand: A Python Cascade

OpenAI shipped GPT-6 Sol and GPT-6 Luna on September 22, and the part everyone is passing around isn't a benchmark. It's the price table. Sol now costs $2 in and $10 out per million tokens, Luna costs $0.10 and $0.50, and both are 50% cheaper than the GPT-5.6 versions they replace. Within hours the OpenAI developer forum had people posting Codex bench runs where Luna passed the same coding tasks as Sol at roughly a tenth of the cost.

That gap, 20x on every token, is what makes a cascade worth building. Instead of picking one model for everything, you send each request to Luna first, check the answer with rules the model can't argue with, and only pay Sol prices when Luna's answer fails those checks. Easy traffic stays cheap. Hard traffic still gets the stronger model.