What changed
OpenAI released GPT-6 Sol and Luna on September 22, cutting prices by half. Sol costs $2 per million input tokens and $10 per million output. Luna costs ten cents and fifty cents. Both are permanent rates, not promotions. The move came 90 minutes after Anthropic cut Claude Opus prices to $4 and $20.
What you can do with it now
Luna handles high-volume work: summaries, extracts, short answers, form fills. At ten cents per million words in, it costs less than one cent per 10,000 words. That is ten proposals, fifty follow-up emails, or two weeks of meeting notes. Build a two-tier setup: let Sol or Astra write the plan, hand Luna the repetitive tasks, bring the result back to Sol for review. You are routing by cost, not by model loyalty.
What to keep in your own hands
Route by task, not by default. A cheap model that redoes work three times costs more than a strong model that gets it right once. Test both on your actual workflow: time to completion, retries needed, total spend. Don't paste customer lists, contracts, or pricing into any model without checking where that data lives after. Keep API keys separate for each task and turn off any model you stop using.