OpenAI: GPT-5.6 Luna

Luna's row lists identically to a much smaller, much older tier. As observed on 28 August 2026, this row lists $0.20per million tokensopenrouter-models, last checked 2026-09-14 input, the same number openai/gpt-5.4-nano lists — $0.20per million tokensopenrouter-models, last checked 2026-09-14 — despite arriving two minor versions later. The match is recent rather than designed: OpenAI cut Luna's price by 80% on 30 July 2026, four weeks before that reading, and the two rows have only lined up since. Nothing else matches: Luna's context window runs 1050000tokensopenrouter-models, last checked 2026-09-14 against Nano's 400000tokensopenrouter-models, last checked 2026-09-14, and Artificial Analysis's coding index, read on 5 September 2026, separates them the same way — 71.4openrouter-models, last checked 2026-09-14 here against Nano's 56.1openrouter-models, last checked 2026-09-14.

Luna is also the cheapest of the three GPT-5.6 models by a wide margin — only its own openai/gpt-5.6-luna-pro sibling matches it. In the same 28 August 2026 reading its $0.20per million tokensopenrouter-models, last checked 2026-09-14 input undercuts Sol's $2.00per million tokensopenrouter-models, last checked 2026-09-14 — the number Terra's row lists too — by a factor of ten. On the intelligence index as it reads on 5 September 2026, Luna's 37.5openrouter-models, last checked 2026-09-14 trails Sol's 47.1openrouter-models, last checked 2026-09-14 by a far smaller proportion than that. Both of those rates are the top listed provider's for their row rather than necessarily OpenAI's own, and two rows need not be headed by the same provider, so the ten-fold gap is one between listings. It shipped inside the same government-requested preview as its two pricier siblings, reaching roughly 20 organisations on 26 June 2026 before its own 2026-07-09source, accessed 2026-08-28 general release.