OpenAI Codex Model Leak Exposes GPT-5.5, Arcanine, Glacier-Alpha, and Heisenberg Codenames

PiunikaWeb·high signal

OpenAI's Codex model picker briefly exposed a list of unreleased internal models including GPT-5.5, oai-2.1, arcanine, glacier-alpha, heisenberg, and GPT-Rosalind. Developers who tested GPT-5.5 before the access was revoked reported it fixed a problem in 3 minutes that had stalled them for 4 hours, with noticeably faster and more token-efficient output. Sam Altman hinted at a possible Thursday announcement, timing the leak to potential imminent release. Multiple codename patterns suggest distinct model families in development.