OpenAI Is Testing 3 Models. Is One GPT-6.5 Sol?

Author: Universe of AI · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: OpenAI Is Testing 3 Models. Is One GPT-6.5 Sol? · Channel: Universe of AI · 11:27 · Watch on YouTube

Topics: AI developer tools, AI content generation, Markets and economy, AI models

Summary

The video discusses three newly surfaced OpenAI API models—Aurora Blue, Aurora Green Alpha, and Aurora Red Alpha—and explains that they appear to be image models being tested on Arena, rather than new coding or general-purpose models. The presenter reviews sample outputs and says the community is disappointed because many expected a stronger model to compete with Anthropic’s offerings. The video also describes perceived competitive pressure from Anthropic, including anticipated model and product updates, and speculates that OpenAI may be delaying releases. It ends with DeepSeek’s 4.1 Flash model identifying itself as an OpenAI model in some tests, which the presenter suggests could point to distillation, while noting the claim is not proof.

Key points

Three models—Aurora Blue, Aurora Green Alpha, and Aurora Red Alpha—appeared in the OpenAI API.
The presenter says the models are image models being tested on Arena, not new Codex or general-purpose models.
The video describes Red Alpha as faster but lower quality than Green Alpha, based on observed tests.
Sample outputs include realistic-looking images, a TikTok-style screen, and pixel art; the presenter says these are not groundbreaking.
The presenter frames the community reaction as disappointment, since many expected a stronger OpenAI model.
Anthropic’s Opus 5.5 is described as a leading model, with Fable 5.5 anticipated in the coming weeks.
The video discusses competition between OpenAI and Anthropic, including updates to Codex and Claude-related products.
DeepSeek 4.1 Flash reportedly identifies itself as ChatGPT or an OpenAI model in some prompts; the presenter speculates about distillation but offers no confirmation.

Chapters

Start	Chapter	Summary
00:00	Three unexpected OpenAI models	The presenter introduces three Aurora models and explains why many initially expected them to be new coding or general-purpose models.
02:36	The models turn out to be image models	The presenter says the Aurora models are being tested as image models on Arena and reviews early sample outputs and community reactions.
06:13	OpenAI’s competitive position	The video discusses dissatisfaction with OpenAI’s pace, compares its position with Anthropic’s, and considers the pressure of upcoming model releases.
09:22	Product updates and the DeepSeek test	The presenter describes Anthropic product-speed updates, then shows DeepSeek 4.1 Flash identifying itself as an OpenAI model and offers a speculative explanation.

Quotes

All of the three models I mentioned, they are all image models. — 02:36

Do we really need another image model right now? — 05:43

It’s a pixel banana. It’s good, but it’s not really like anything groundbreaking — 06:46

If you remove the default system prompt on open router and you ask this model what model you are, it says it’s an open AI model. — 10:57

Concepts and entities

Aurora Blue, Aurora Green Alpha, Aurora Red Alpha, OpenAI, Arena, Codex, Anthropic, Opus 5.5, Fable 5.5, Claude Code, DeepSeek V4.1 Flash

Full transcript (original language)

00:00 Something interesting happened last night. Three new OpenAI models appeared in the OpenAI API: Aurora Blue, Aurora Green Alpha, Aurora Red Alpha. And all of these models were expected to be new models from OpenAI that were going to improve their Codex platforms or just be new base models for new AI models. But that is exactly what did not happen. So, let’s get into it. There was an open router model called Aurora Alpha earlier this year, which was linked to GPT 5.3

00:30 Codex Spark. So, when people actually saw this model appear, a lot of people were getting excited because it seemed like OpenAI is due for a new model drop. If you look at the current landscape, their models are not competing with Anthropic’s model at the moment. Are they still competitive? Are they still, you know, really great to use? Of course. But if we just look at the AI space, the AI space is very, very competitive. And a lot of people are loving the Opus 5.5 model and all the new models from the Anthropic team versus OpenAI. After their dev day,

01:01 things have just gone a little bit downhill for them because all of their updates that they have been dropping with the Codex updates and things like that, they are good for usability uses. But when it comes to having the most competitive model, that is not happening. So, when people actually started to see this, a lot of us were excited, including myself, because this indicated that OpenAI could be dropping a specialized Codex coding models. They have done that in the past. And if we look at what’s happening at the moment, how they’re improving the Codex features all over the next 28 days, it could make

01:33 sense for them to drop a model fine-tuned for the Codex experience or a high-speed inference version of the GPT-6 models. Some people are even expecting that the new models from OpenAI would be able to compete against Opus 5.5 because obviously we know that this is probably the best model at the moment. The Anthropic model lineup usually goes by Fable, Opus, Sonnet, and Haiku. And currently Opus 5.5 is the strongest model. And over the next few weeks, we are expecting Anthropic to actually drop a new model as well, Fable

02:04 5.5. And because of all that pressure that has been building, we were expecting OpenAI to drop something new. However, disappointingly, hey, before we get into the video, I just want to take a moment to tell you guys that AI is not slowing down anytime soon. And if you guys want to stay on top of everything that is happening in AI, be on top of all the new model drops, then make sure you guys are subscribed to the channel. Your support means a lot. And now, let’s get back into the video. All of the three models I mentioned, they are all image models. Yes, they

02:36 weren’t going to be the next generation of Astra, Soul, or Luna models, but they were all actually just image models that are being tested in Arena right now. And like this person, this was a big disappointment. Because from OpenAI, we were expecting newer models from them. They do have a model named Bell that it’s in works internally, probably being used. We just don’t know when we were getting it. And when people saw this, it didn’t make sense for a lot of us because image models, like we don’t really need new image models at the

03:07 moment. Obviously, new AI development is always appreciated. But from OpenAI, a lot of us weren’t waiting for them to drop a new image model. We’re waiting for them to drop a new stronger version of Astra, Soul, or even Luna. Because GPT 6.1 Soul, although it is more efficient than the Soul previous version, it’s still not a competitive model when we compare against Opus 5.5. So, like this person said, a lot of us are probably going to be disappointed because all of us got excited yesterday night thinking that we were getting a

03:38 new model from OpenAI, but it’s not going to be a new model. Most likely, it’s going to be a new image model. And we don’t know when this new model is going to be dropping. We do know that it’s being tested on Arena, and I’m going to show you guys a couple of outputs from that model, but it’s just disappointing to see what OpenAI is doing at the moment. And at the end of this video, I’m actually going to show you something interesting. Yesterday, we also got a new model from DeepSeek. Now, what’s interesting, this model is actually saying that it’s an OpenAI model. So, stay tuned until the end of

04:09 the video to see that as well. So, the first test that you’re seeing is that obviously people realized that these new models that we’re getting from them are going to be image models. And the overall what we have noticed is that Aurora Red Alpha is the faster model out of those three, but it is producing not the best quality of work. Then we have Green Alpha, which is a little bit slower and better quality, which does make sense. You Maybe the model is thinking a little bit harder, and that is why we are expecting that model to produce a better quality output compared

04:41 to a faster model. And obviously, this is on Arena, so you either get lucked out when you get this new model or you don’t. And it looks like this person didn’t luck out, but somebody did, and I’ll show you the blue model as well. But this is an output. Yeah, it’s it’s good. It’s you know, really looks realistic, things like that. But other than that, like, you know, it’s not going to be a major update from GPT image 2, cuz that model is obviously very competitive already. So, here’s another person that was testing this, and he seems to be obviously happy that

05:13 OpenAI is testing a new image model. This is the result that we see at the bottom over here. This result is from Red Alpha. And this does look like an actual TikTok screen. Do the people look a little bit AI? Yeah, but like if you just saw this, you would just think that this was a screenshot from TikTok rather than, you know, obviously, an image model. There’s some errors over here if you look at the bottom. I don’t think that number should be there, like the 1987 versus like everything else over here. As this person also says, “Do we really need another image model right

05:43 now?” Probably not, and that is why a lot of us are probably a little bit disappointed. A lot of us probably were excited that this was going to be a new base model for upcoming open AI models, but that is not the case and maybe those base models would have been tested on Arena and we would have gotten some insight onto how powerful Bella is. Now, one thing to understand is that OpenAI currently is dealing with a problem with their models where they have been escaping their sandbox testing environments and like, you know, doing things that they’re not supposed to do.

06:13 And that is probably why they have delayed their drops a little. But to be fair with Anthropic, they also have internal models that are stronger than the ones out there, but they seem to be able to keep up with the pace of shipping, which is ironic because they were the ones originally that said that we need to pace the frontier. But it looks like OpenAI is pacing the frontier and not Anthropic anymore. Interesting times, I’m telling you guys. But yeah, this is crazy because we might be getting new image models from OpenAI and not, you know, stronger models from them yet. Here’s another example, which looks

06:46 like a pixel art and apparently this was something that I personally haven’t tested OpenAI image models a lot, but they didn’t do really well at pixel art. But it looks like the Aurora Red Alpha has done a good job of creating a pixel banana. Um looks like a pixel banana to me, but I I like you guys can see from my tone I’m not like that amazed by it. Like, okay, it’s a pixel banana. It’s good, but it’s not really like anything groundbreaking, you know what I mean? And yeah, across the AI community, the overall sentiment that people are

07:17 feeling right now is what exactly is OpenAI doing? They obviously, as this person said, they fumbled the GPT-6 so launch and then after Anthropic dropped Opus 5.5 very close to the GPT-6.1 launch, they have been just killing it. They have been making improvements to their models, they have been dropping more efficient models, and they’ve also been reducing the prices for their models, and they have been able to attract a lot of new users to their platform. So, we can see from the growth that they are experiencing. And now when people saw these mysterious Aurora

07:48 models, a lot of us were excited that we were thinking that we were going to get a new model from them. But, yeah, it turns out that that’s an image model. So, yeah, it looks like the excitement at OpenAI is not really good. The sentiment around OpenAI is, “What are they doing? What’s taking them so long?” That’s what people have been feeling. But, that’s the AI space because realistically the newest model from OpenAI literally dropped a few weeks back. It hasn’t been that long. But, since we are expecting new models from these labs very rapidly now, especially in 2026, a lot of people are

08:20 disappointed, which is understandable. And because you are paying a lot of money to have these subscriptions, you are paying a lot of money to run these models. And, you know, you want your money’s worth. And that is why a lot of people are disappointed. And as I said, Anthropic is not slowing down anytime soon, either. Fable 5.5 is expected to come within a few weeks. A lot of people, you know, are waiting for this model. We are seeing that this model Fable 5.1 is being routed to a new stronger Fable 5.5 model and producing outputs that are genuinely going to put

08:52 more pressure on OpenAI if they have no answer to this model. Because this is an SVG test from the Fable 5.5, which pretty much tells you that this test is going to be retired soon. But, SVG code and just overall outputs that we have been seeing from Fable 5.5 are quite mind-blowing. And if this model comes out and OpenAI still hasn’t dropped something, things are going to get a little bit tougher for OpenAI. And remember, earlier this year OpenAI had to ring code red, if you guys remember that. And that code red was what led to the new Codex update, newer models,

09:22 things like that. So, it feels like that could be happening again. It could be OpenAI’s time to ring in another code red because at the moment they do not have an answer to any of the newest models. And what’s interesting, as I said, that over this next 28 days open AI has been working on making Codex updates, but Anthropic is also making some Cloud code updates. For example, they have made the whole experience in Cloud code three times faster and 1.5 times faster in Cloud. If you look at the local sessions in Cloud code, what they have basically

09:54 done is that they have started to take your conversations and turn them into G zips and because of that they have achieved some faster responses and we can see Cloud code local sessions three times faster. Cloud sessions are 2.8 times faster. The chat and co-work features are now 1.5 times faster and the API is 1.5 times faster. So yeah, it looks like Anthropic is also updating their platforms like Cloud code, not just Codex. Now this is a funny thing that I want to show you guys. So yesterday Deep Seek team dropped version

10:24 4.1 flash model and you know, it’s a very efficient model. It’s much cheaper than any of the frontier labs and it’s producing outputs that are very competitive. But when you ask this model what model you are, it often says that it is a chat GPT model and AI assistant developed by open AI. You can see it’s been asked this question multiple times and every time it says an AI assistant. And then people have tested this by themselves and it actually does say it’s chat GPT. If you remove the default system prompt on open router and you ask this model what model you are, it says

10:57 it’s an open AI model. So maybe open AI secretly has been working for the Deep Seek team. Obviously, I’m joking. This just indicates that maybe the Deep Seek lab distilled their model from open AI’s models, but regardless, I hope open AI drop something new because I personally love the open AI ecosystem. They’re obviously going to drop something new and you know, the then the whole sentiment once again is going to shift. We notice that’s how the AI space has always been, but let’s see when that happens. But that’s it for today’s video. Make

11:27 sure you guys are subscribed to the channel. Follow our new newsletter as well at universeofai.beehive.com as well subscribe to the main channel World of AI and support us on X by following the universeofai.z as well. Until then, I’ll see you guys in the next video.