It's the first time in the history of the world.
Recently, three new models of OpenAI have been briefly launched.
Recently, the OpenAI API backdrop has unveiled three mysterious alpha models called aurora-red-alpha, aurora-green-alpha, aurora-blue-alpha.
Do we want to introduce a new model again?
However, before everyone was ready to test it, OpenAI immediately withdrew the three models!
However, while the API was blocked, the three models completely exposed their traces in the Arena.
The results of the tests were surprising.
This isn't a code model, it's the baseline for the next generation of King-Break images that OpenAI is about to release!
The three models were suddenly new and withdrawn late at night.Yesterday, everything happened too fast.
When three models such as aurora-red-alpha appeared on the API list, the developers were thrilled.
Some speculate that this may be the legendary GPT-6.1 Astra.
Earlier in October, OpenAI suspended GPT-6.1 Astra due to security concerns.
And on February 9 of this year, OpenRouter suddenly installed one.「Aurora Alpha」Anonymous model.
The official description reads: "This is an intangible model for collecting feedback from the community, a reasoning model created specifically for speed".
Then, just three days later, OpenAI released GPT-5.3-Codex-Spark in high definition.
OpenAI claims it is a "super-fast model for real-time coding in codex", generating more than 1,000 tokens per second.
Because the release date is so close, the performance labels are so consistent. Since then, the community has defaulted to an equation: Aurora = OpenAI's ultra-high-speed/strongest code model.
The word "Aurora" stands for top-level code and reasoning in everyone's mind.
Just when everyone thought these three "polar light" models would give off ultra-high-quality code, someone in the Arena discovered their true faces.
They are not super-high-speed code generators or variants of the GPT-6.1 Astra.
These three high-altitude models are generating images in the Arena.
In fact, I'm painting cats.
As it turns out, while the APIs were closed, OpenAI did not remove them from Arena's grayness test pool to collect RLHF data.
All the netizens immediately turned to the Arena, launching a wave of extreme testing.
In an exciting mood, the testers sent Prompt to the aurora-red-alpha.
The result came back, aSuper-sophisticated cat faces。
The image is extremely rigorous, photographically sophisticated, and gives a sense of "intentional and professional people like photography".
They then tested the aurora-green-alpha again.
This time it's still a cat.
But unlike the red polar beam, the green polar beam produces a single beam of light.A wider view, a brighter backdrop, and a lighter interior.It's like a hand-painted moment of life, natural and arbitrary.
As for the aurora-blue-alpha, although no specific cat form has yet been published, it is certain that it is also an image-generating model.
She was silent for three seconds and then laughed silently.
We thought OpenAI was going to rule the world, so Altman just wanted us to paint a different style of cat!
In Arena's blind test, users are required to vote for two models.
We soon found out:The Red version of Aurora-Red is being overwhelmed by a huge margin of victory.
This red-and-green battle is actually a key confrontation between OpenAI and the next generation of Image to define the default graphic style.
So far, the "green version", which represents aesthetic excellence and natural taste, has won the prize.
So what about the blue version of Polar Light?
After hours of searching, they finally found Aurora Blue Alpha!
After that, more details about the output of the blue version!
Some people tested the blue version of the "Bicycle Ride" and it was magical.
The comics look good.
It has been said that the images created with Aurora look like the result of a collaboration between Nano Banana and GPT Image.
According to the results of the demo, Nano Banana 3 Pro will face fierce competition.
I'm sorry, this could be the case today.
A "mistake" in OpenAI's API exposed three models of ultraviolet light named RGB triangles.
They could be Image-3, and they could be the visual underpinnings of OpenAI's next generation of multi-mode engines.
Of course, OpenAI's ambition is not just to generate images.
Recently, they have had a series of intensive moves.
GPT-6 with Intelligent UI has just made its debut in ChatGPT.
Dots, the flagship product designed to replace Grok and Muse, has come online, backed by GPT-6 Astra and GPT-6.1 Sol for computational support.
Just last week, the pricing page was crammed into an unpublished model called gpt-rosalind-discovery, and immediately after that the Rosalind Workbench was opened to the public as a research preview!
Now, the Anthropic killer Fable 5.5 is approaching.
Everyone is hotly debating the next generation of generic models, resulting in only image models.
Whether it's the visual evolution of GPT-Image-3 or looking at a larger cutting edge model, OpenAI will have to come up with something hard enough.