GPT-5.6 Sol, Terra, and Luna: How to Choose
A practical framework for choosing among GPT-5.6 Sol, Terra, and Luna based on task difficulty, total cost, latency, and review effort.
Affiliate disclosure: This article may later contain clearly labeled affiliate links. Our reporting and conclusions are not sold. Read the full policy.
The short answer
OpenAI’s GPT-5.6 family uses three product positions: Sol for the most demanding work, Terra for balanced everyday tasks, and Luna for cost-efficient volume. Choose by completed-work economics, not by defaulting every request to the strongest model.
What changed
OpenAI launched the family across ChatGPT, Codex, and the API, then announced pricing changes during the release period. That is a reminder that model routing needs configuration rather than hard-coded assumptions.
The names describe intended tradeoffs, but the correct route depends on your prompts, tools, output length, and acceptable error rate.
Why it matters
A cheaper model can be expensive if it creates more retries or human corrections. A flagship model can also waste budget on extraction, classification, or formatting tasks that a smaller option completes reliably.
Total cost should include input and output tokens, tool calls, latency, failed runs, and reviewer time.
A routing method
Build a task set with three bands. The first band contains deterministic work such as field extraction and format conversion. The second contains analysis with known source material. The third contains ambiguous, long-running work with several constraints.
Run all three models on each task. Set an acceptance rule before viewing results. Route a task to the least expensive model that passes the rule consistently, and escalate only after a failure or when risk justifies it.
Recheck over time
Model behavior, prices, context limits, and tool support change. Store model identifiers and evaluation dates with results. Re-run the suite after a model update or a major prompt change.
The decision
Use Luna for proven high-volume routines, Terra as the general starting point, and Sol for tasks where added capability reduces enough failure or review to justify its cost. Treat that as a hypothesis to test, not a permanent hierarchy.
Primary source: OpenAI model announcement. Last reviewed September 11, 2026.