The cheapest GPT-5.6 tier; classification, routing, extraction, high-volume chat.
Luna is the efficiency tier of the GPT-5.6 family and the most striking price story in the survey: an 80% cut on July 30, 2026 — the steepest single reduction of the year from a US lab — brought it to $0.20 / $1.20 per million tokens. Crucially, it retains the family's full 1.05M-token context.
Luna is not built for the deepest reasoning, and it has no standalone independent intelligence score here; the long-context surcharge still applies at scale. Those are the expected trade-offs for a model priced this low.
Where it excels is high-volume, well-scoped work: classification, routing, extraction, and large-scale chat, where the enormous context and rock-bottom price matter more than frontier reasoning. For that class of task Luna is one of the most cost-effective closed models available; for hard reasoning, step up to Terra or Sol.
FIELD NOTE — Cut 80% on Jul 30 — the steepest single cut of the year from a US lab.
VISIT OFFICIAL SITE →