Three model identifiers - aurora-red-alpha, aurora-green-alpha and aurora-blue-alpha - were reported in OpenAI's API on 18 October and spread quickly through X. We are covering this with the caveat attached rather than buried: the original sighting is unverified, no outlet has independently reproduced it, and OpenAI has said nothing.
What can be stated plainly: there is no OpenAI announcement, no documentation, no pricing, no rate limits, no stated context length and no changelog entry for any of the three. The colour suffixes have been read as hints at Codex variants or faster inference tiers, but that is inference from three words in a model string, not information. An earlier identifier, aurora-alpha, surfaced on OpenRouter in February 2026 and has been speculatively tied to a Codex variant, also without confirmation.
Unannounced IDs showing up in an API is routine - labs stage models behind access controls before launch, and alpha suffixes usually mean exactly that. The useful takeaway is not that OpenAI has three new models. It is that a single X post about a model string can become a news cycle in a day, and that the sensible response is to wait for the changelog.