The OpenAI Aurora models are three unannounced model names that briefly appeared in OpenAI’s API on October 18, 2026. They look like image generators, and the community guess is a “GPT Image 3,” but OpenAI hasn’t confirmed any of it.
So this is a leak story, not a launch. Here’s what is solid, what is guesswork, and what you should (and shouldn’t) do about it.
A quick snapshot first.
- Names spotted: aurora-red-alpha, aurora-green-alpha, aurora-blue-alpha
- When: October 18, 2026
- Likely type: image generation (unconfirmed)
- Official docs, pricing, or release date: none
- Can you use them? Not reliably; they were short-lived and undocumented
The takeaway: treat this as an early signal, not a product you can plan around.
What exactly showed up in the API?
According to Runtime Wire, the identifiers were first shared on X by a user known as Lisan al Gaib. The post didn’t include a working API response or any model metadata.
That matters. A name in a list isn’t proof a model works, and nobody has shown a successful request. OpenAI hasn’t said anything publicly.
TestingCatalog reports the names appeared for a short time over the weekend and were pulled soon after. It suggests the rollout may have been accidental, but that’s its read, not a statement from OpenAI.
Why people think it’s GPT Image 3
Within hours, three models appeared in blind image-generation tests on Arena. Lumina Bench said all three appear to be image models, and guessed they belong to a new GPT Image release.
The early impressions are mixed and thin. Red reportedly generated faster but at lower quality, green was slower but looked better, and blue hadn’t been tested yet. Two sample cat pictures say very little about real quality.
Also worth knowing: three other unnamed models (Aspen, Cypress, and Maple) showed up on Design Arena about an hour later. Nothing links them to Aurora, so don’t merge the two stories.
Fact vs. guess: a quick sorting table
With a leak, it helps to keep the solid facts apart from the rumors. This table does that in one glance.
| What we know | What we don’t |
|---|---|
| Three IDs showed up: aurora-red-alpha, aurora-green-alpha, aurora-blue-alpha | What each one actually does |
| They briefly appeared in OpenAI’s API on October 18 | Whether they will ever be public |
| Matching unnamed models showed up in blind Arena tests | That those tests are the same models |
| OpenAI has not published docs or pricing | Release date, price, or the final name |
If a headline claims a price or date for these models, it hasn’t come from OpenAI. Be skeptical.
Why AI labs test models in public
Blind arenas are a cheap way to see how real people rate a model before launch. Voters don’t know which lab made what, so the feedback is less biased than a company’s own benchmarks.
TestingCatalog notes OpenAI has tested image models this way before. That history is why many people read these names as a real pre-release test rather than a random glitch.
There’s a catch, though. Arena builds are often early versions, so speed and quality you see in a test may not match what ships. A model that looks rough today could improve a lot by launch, or get shelved entirely.
That’s also why a single leak can’t tell you much about the final product. Wait for the official release notes, then judge it yourself on your own prompts.
What this means for developers and ChatGPT users
If you build on the OpenAI API, don’t code against these names. Undocumented model IDs can change, vanish, or return errors without warning. Red, for instance, was reported to return capacity errors, while green and blue returned not-found errors.
If you just use ChatGPT, nothing changes today. OpenAI has been busy elsewhere lately, from the interface changes in our GPT-6 Intelligent UI rundown to the rollout covered in our ChatGPT Slides guide. A new image model would fit that pattern of steady, quiet expansion.
The practical move is to watch OpenAI’s official changelog and model list. When the models are real, they’ll show up there with pricing and limits.
How to follow the OpenAI Aurora models without the noise
Stick to a few reliable habits. They save you from acting on rumors that fall apart by tomorrow.
- Check OpenAI’s own announcements and API model list before believing any screenshot.
- Treat Arena results as hints; blind tests are small and models are often tuned before launch.
- Wait for pricing and rate limits before you estimate any project cost.
Leaks like this often turn into real launches, but timing is the part nobody outside OpenAI knows.
Frequently Asked Questions
Are the OpenAI Aurora models released?
No. The names appeared briefly in the API, but OpenAI hasn’t announced, documented, or priced them.
Is Aurora the same as GPT Image 3?
Unknown. Testers think the models are image generators and guess at that name, but it’s speculation.
Can I call the aurora models through the API?
Not in any dependable way. Reports say they were removed quickly and returned errors for some IDs.
When will OpenAI launch a new image model?
There’s no official date. Watch OpenAI’s changelog for a confirmed announcement.
My take: this looks like a real image-model test, and a launch in the coming weeks wouldn’t surprise me. But until OpenAI posts docs and pricing, keep your production code on documented models and just enjoy the sneak peek.
Leave a Reply