GPT-6 Luna

GPT-6 Luna is the high-volume model of OpenAI’s GPT-6 family, announced on 22 September 2026 and served as gpt-6-luna. Its documentation page gives the same 1,050,000-token context window and 128,000-token output limit as GPT-6 Sol, text and image input with text output, and a knowledge cutoff of 18 May 2026. It is priced at $0.10 per million input tokens, $0.50 per million output tokens and $0.01 per million cached input tokens — half the $0.20 and $1.20 charged for GPT-5.6 Luna. OpenAI positions it for focused, high-frequency work rather than long reasoning, and says that at high effort it improves on its predecessor by 5.4 percentage points at 58 per cent lower cost per task; that is OpenAI’s own measurement and is not stored as a score. Free and Go users reach it in the ChatGPT desktop app; Plus, Pro, Business, Enterprise and Edu users reach it in ChatGPT Work and Codex. Artificial Analysis, measuring independently, scores it 37 on version 4.3.2 of its Intelligence Index, records 157.2 output tokens per second — the fastest of the three frontier models released on 22 September — and puts its blended rate at $0.08 per million tokens on a 7:2:1 cache-hit, input, output mix. The gap between Luna and Sol on that index is eleven points for a twentyfold difference in the input price, which is the trade to test on your own task rather than to accept from a chart. As with every model on OpenAI’s API, the lifecycle is published: the deprecations page lists dated shutdowns, and the twelve legacy snapshots announced in April 2026 stop answering on 23 October 2026.