GPT 5.6 Sol vs Terra vs Luna (URGENT)

OpenAI shipped GPT 5.6 in three tiers: Sol at $5 in and $30 out per million tokens, Terra at $2.50 and $15, and Luna at $1 and $6. So I gave all three the exact same build job on one $20 subscription, with the serving model verified in every chat. All three built a working, math correct tool. The $1 model finished first. Here is what checking the receipts turned up. The pages where you actually decide what to pay, the model cards, the compare page, even GitHub's announcement, publish zero benchmark numbers. The real per tier scores are buried in the system card safety PDF. Independent testing puts the coding gap between the $5 tier and the $1 tier at about 6 percent, while the price gap is 5 times. And the warning label nobody reads: OpenAI's own card documents the model taking actions you did not ask for, and the only independent evaluation flagged the highest benchmark gaming rate the lab has ever recorded, with no retest since launch. Match the model to the task. Cheap models for the labor, expensive models for the decisions. Every number here comes from OpenAI's own documents, independent evaluators, or my own test. ▶ Subscribe for receipts, not hype:    / @predir_ai   ⏱ CHAPTERS 0:00 One sub, three models 0:11 The $1 model finished first 0:22 The missing numbers 0:45 Four things 1:05 The real prices 1:18 Vibes, no numbers 1:30 The system card numbers 1:52 The independent numbers 2:08 6% vs 500% 2:26 My three way test 2:49 What each tier built 3:14 The hidden tiers 3:30 The warning label 4:03 No retest since launch 4:14 The verdict 4:44 Every number sourced 📌 Sources GPT 5.6 System Card (per tier benchmark tables): https://deploymentsafety.openai.com/g... OpenAI model docs and pricing (Sol, Terra, Luna): https://developers.openai.com/api/doc... Artificial Analysis independent benchmarks: https://artificialanalysis.ai/article... METR predeployment evaluation of GPT 5.6 Sol: https://metr.org/blog/2026-06-26-gpt-... GitHub Copilot changelog (tier positioning): https://github.blog/changelog/2026-07... This is Predir AI. Build real things with AI. #GPT56 #OpenAI #Sol #Terra #Luna #AIComparison