OpenAI's GPT-5.6 family began rolling out from late June 2026, and it doubles down on a trend we've covered: frontier models ship as families, not single models. GPT-5.6 comes in three tiers — Luna, Terra, and Sol — all sharing a 1-million-token context window and a February 2026 knowledge cutoff, priced from roughly $1 to $5 per million input tokens.
The three tiers
- Sol — the flagship, top capability, with a new "Ultra" subagent mode and a "Max" reasoning-effort setting for the hardest problems.
- Terra — the balanced middle, targeting the previous generation's flagship quality at roughly half the cost.
- Luna — the fast, cheap tier for high-volume, latency-sensitive work.
One model never fit every task. A tiered family lets you route the easy majority to the cheap tier and reserve the flagship for the hard minority — which is how you control cost.
Why the family model matters
This is the pattern the whole industry is converging on (Gemini and others do the same): match the model to the request. Sending every query to the flagship is like couriering every letter by private jet. With a family, you develop against one tier and scale up or down without re-engineering, because they behave consistently.
The practical takeaway
If you're building on GPT-5.6, think in terms of routing, not a single choice: Luna for common, simple, high-volume calls; Terra for the balanced default; Sol (with Ultra/Max) for genuinely hard reasoning or agentic runs. As always, benchmark on your own tasks — and remember that the cheapest tier clearing your quality bar, not the flagship, is usually the right default at scale.
