Three tiers within one generation
Sol, Terra, and Luna are the three model tiers introduced with OpenAI’s GPT-5.6. Separately from the generation number, they express differences in capability, speed, and cost.
Rather than routing everything through one enormous model, the tiers let you match the model to what the work actually requires. OpenAI positions them as capability tiers that persist across generations.
Sol: the top tier
Sol handles complex reasoning, long-running agentic work, large-scale coding, and specialist analysis. It shows its advantage on work with many constraints that requires repeated investigation and judgment along the way.
Used on simple tasks it consumes disproportionate time and allowance. It rewards recognizing what is actually difficult.
Terra: the everyday choice
Terra balances quality, speed, and cost. It suits writing, research, ordinary coding, and organizing material — the bulk of daily work.
Where the procedure and the desired result are reasonably clear, Terra is often enough. Starting with Terra and escalating only the hard parts to Sol works well.
Luna: speed and volume
Luna is the fast, low-cost tier. It suits formatting, classification, extraction, and boilerplate — the same kind of processing repeated at scale.
The three are a division of labor rather than a ranking. Sol for difficult work, Terra for everyday work, Luna for fast bulk processing is the summary worth remembering. Availability and pricing per tier can change, so check current information before relying on it.