OpenAI exposes three Aurora models in its API: red, green and blue alpha
On October 18, developers spotted three unfamiliar identifiers in the OpenAI API model list: aurora-red-alpha, aurora-green-alpha and aurora-blue-alpha. All three are named after Aurora, differ only by color and carry the suffix alpha. OpenAI has said nothing publicly about them: not the architecture, not the purpose, not the release timing. This is not the first time the company has revealed code names through technical interfaces ahead of announcements. • Three identifiers: red, green, blue. • The alpha suffix points to an early testing stage. • There are no official specifications or confirmation yet.
AI-processed from Jiqizhixin (机器之心); edited by Hamidun News
Three new identifiers in the OpenAI API model list, discovered on October 18, became the biggest find of the weekend for people who follow the company. Developers saw aurora-red-alpha, aurora-green-alpha and aurora-blue-alpha. These names do not yet appear in OpenAI's documentation, blog or statements. The discovery was reported by Jiqizhixin.
What exactly was found
The facts fit in a few lines. The API output that lists available models now includes three unfamiliar names. They share the root Aurora (in Russian, «полярное сияние», or "polar lights") and differ by color: red, green and blue. The suffix alpha usually denotes an early, non-public stage, when a model is tested on a limited set of accounts or inside the company's infrastructure.
It is still unknown who can access these models, whether they respond to requests, and what a call actually returns. There are no confirmed benchmarks, context window sizes or prices.
What the names might tell us
What follows is speculation only, and it is important not to present it as knowledge.
A color scheme under a shared name looks like three variants of one model line. They could be different sizes, different operating modes (fast, reasoning, multimodal) or three independent experiments under one code name. An A/B test of different checkpoints of a single model is also possible. The names alone do not reveal which of these scenarios is correct.
There is precedent for this practice. In 2024, anonymous models such as gpt2-chatbot appeared on LMArena and were later linked to OpenAI. Labs use code names in public interfaces to collect real-world feedback before an announcement. But an identifier in the API list may also be a technical artifact: a test entry that was exposed by mistake and will soon be removed.
What is known and unknown
- Date of discovery: October 18, 2026.
- Number of models: 3.
- Identifiers: aurora-red-alpha, aurora-green-alpha, aurora-blue-alpha.
- Stage, judging by the suffix: alpha, meaning early.
- Official confirmation from OpenAI: none.
- Architecture, context, modalities, prices, timing: not disclosed.
How to treat the find
Developers should not build product plans around models that are absent from the documentation. Alpha identifiers can disappear without warning, change behavior or never reach release. If you have API access, you can check your own model list, but it is better not to build production integrations on such names.
Those who follow the market are better off waiting for confirmation: an official announcement, an entry in the documentation or results on independent benchmarks. The mere appearance of three names says only that OpenAI is working internally on several variants of something new. What exactly that is, we do not yet know. We will return to the topic when verifiable data appears.
- Oct 24, 2026Math Advisers Say OpenAI's Flood of Proofs Broke the Field's Rules
- Oct 24, 2026Vercel Takes Python to GA and Launches Daemons for Background Workers
- Oct 24, 2026SpaceXSI Bundles X Premium, Grok and Cursor into a Single $8–$200 Subscription
- Oct 24, 2026OpenAI's Revenue Is $50 Billion, Not $70 Billion, and the Company Is Seeking $30 Billion
❓ Key Questions Answered
What exactly was found?
The facts fit in a few lines. The API output that lists available models now includes three unfamiliar names. They share the root Aurora (in Russian, «полярное сияние», or "polar lights") and differ by color: red, green and blue. The suffix alpha usually denotes an early, non-public stage, when a model is tested on a limited set of accounts or inside the company's infrastructure. It is still unknown who can access these models, whether they respond to requests, and what a call actually returns. There are no confirmed benchmarks, context window sizes or prices.
What the names might tell us?
What follows is speculation only, and it is important not to present it as knowledge. A color scheme under a shared name looks like three variants of one model line. They could be different sizes, different operating modes (fast, reasoning, multimodal) or three independent experiments under one code name. An A/B test of different checkpoints of a single model is also possible. The names alone do not reveal which of these scenarios is correct. There is precedent for this practice. In 2024, anonymous models such as gpt2-chatbot appeared on LMArena and were later linked to Ope
What is known and unknown?
Date of discovery: October 18, 2026. Number of models: 3. Identifiers: aurora-red-alpha, aurora-green-alpha, aurora-blue-alpha. Stage, judging by the suffix: alpha, meaning early. Official confirmation from OpenAI: none. Architecture, context, modalities, prices, timing: not disclosed.
Build & Ship Production AI Apps with Vibe Coding
Join Zhemal Khamidun's hands-on Vibe Coding Intensive at hamidun.com/vibecoding — master Claude Code, Cursor, and autonomous agentic workflows.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.