GPT 5.6 exposes a configuration matrix of Work/Codex × Sol/Terra/Luna × six reasoning-effort levels × Standard/Fast, yielding 72 possible combinations. The post frames these choices through the lens of training-time vs. inference-time scaling, noting that model size (Sol, Terra, Luna) maps to training compute while effort settings map to inference compute. The author raises the practical question of which combination offers the best performance-to-cost tradeoff, without providing a definitive answer.
7 Impressions