GPT-6 Luna

Published frontier peer. Closed API gpt-6-luna. Only banked cite: DeepSWE v1.1 at max 66.6% from the OpenAI launch page. The page’s AutomationBench line is a relative +5.4 percentage-point claim, not an absolute score, so AutomationBench stays empty. Terminal-Bench 2.1 and SWE-bench Verified are empty. Not a LocalsOnly desk run. https://openai.com/index/introducing-gpt-6-sol-and-luna/