GPT-6 Sol/Luna vs. Opus 5.5: A Real Price War Breaks Out

OpenAI shipped GPT-6 Sol and Luna yesterday — two new models positioned explicitly around the capability-vs-cost tradeoff — and within hours Anthropic countered with Opus 5.5 at 40% less than Opus 5. Simon Willison is calling it a new price war, and he's right. Meanwhile, Shreya Shankar posted a sharp mental model for why LLM apps fail in production that every PM building on AI should read, and Pieter Levels crossed $10M/year at a 94.5% margin — a useful reminder of what's possible when you build lean on this infrastructure.