OpenAI’s GPT-6 Luna launched at $0.10 per million input tokens and $0.50 per million output tokens; GPT-6 Sol is $2 and $10. Simon Willison’s September 22 comparison puts both at roughly half the promotional price of their GPT-5.6 counterparts. Anthropic also cut Claude Opus 5.5 to $4 input and $20 output, with cache reads down 60%. Pricing comparison and first tests
For agent economics, the cache price may matter more than the headline input price: long tool-using sessions repeatedly reread their own history. Lower cost also changes architecture. A team can route routine classification to Luna or a decision model such as Jev, reserve Sol or Opus for ambiguous work and still afford more verification calls. Early capability claims remain unsettled. Willison twice watched Opus 5.5 exhaust its 128,000-token reasoning allowance while drawing a pelican on a bicycle—an expensive reminder that more thinking is not a monotone good.