Claude Fable 5.2 and Opus 5.2 Are Coming Sooner Than Expected!

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: Claude Fable 5.2 and Opus 5.2 Are Coming Sooner Than Expected! · Channel: Universe of AI · 10:55 · Watch on YouTube

Topics: Claude Code, AI content generation, Markets and economy, AI models

Summary

The video presents claims that Claude models referred to as Fable 5.2 and Opus 5.2 are being secretly tested, and describes a question about an OpenAI employee as a way some users believe they can identify the newer routing. It compares reported outputs from the models with earlier models and GPT-6 Astra on image-generation tasks, including a controller, a pelican riding a bike, rockets, and 3D environments. The presenter says the examples look improved, especially for realistic 3D generation, while noting that efficiency, speed, and pricing are uncertain. The video speculates that the models may arrive closer to November and discusses how OpenAI’s upcoming developer event and competition could affect Anthropic’s plans.

Key points

The presenter says Fable 5.2 and Opus 5.2 may be secretly routed in Claude Code.
A question about Tibo, an OpenAI Codex team employee, is described as an informal test for possible routing to a newer model.
A PS5 controller comparison is presented as an example of Fable 5.2 producing a more recognizable result than Fable 5.1.
The video compares Opus and Fable outputs with GPT-6 Astra on a pelican-riding-a-bike prompt, while acknowledging that the test may no longer distinguish models well.
Rocket and 3D environment examples are used to highlight more realistic, coherent generations.
The presenter cautions that these are visual demonstrations, not evidence about efficiency or pricing; Anthropic models are characterized as potentially slower and more expensive.
The presenter speculates that Fable 5.2 or Opus 5.2 could launch closer to November, linking this possibility to Anthropic’s reported IPO plans.
OpenAI’s upcoming developer event and competition with Anthropic and Google DeepMind are framed as important to the coming quarter.

Chapters

Start	Chapter	Summary
00:00	Possible secret testing of Fable 5.2	The presenter introduces claims that Fable 5.2 and Opus 5.2 are being tested and describes an informal question-based test for identifying newer model routing.
02:06	Comparing model-generated images	Examples including a PS5 controller, a pelican riding a bike, and rocket generations are used to compare the supposed newer models with earlier outputs and GPT-6 Astra.
05:13	3D generation, quality, and trade-offs	The video highlights realistic-looking 3D and game-like generations while noting that speed, efficiency, and pricing remain uncertain.
08:18	Release timing and competitive pressure	The presenter speculates about a November release and discusses IPO timing, OpenAI’s developer event, and competition among AI labs.

Quotes

So, looks like Fable 5.2 is secretly being tested in cloud code. — 00:00

If your model is able to answer this particular question, then you know it’s secretly being routed to Fable 5.2. — 01:03

I think in my opinion, it’s time to retire this test — 04:11

We don’t know what the pricing is going to be for Fable 5.2 or how efficient this model is going to be. — 05:45

could be releasing closer to the November timeline — 08:52

Concepts and entities

Fable 5.2, Opus 5.2, Claude Code, Anthropic, OpenAI, GPT-6 Astra, Codex, Tibo, Roblox Studio

Full transcript (original language)

00:00 So, looks like Fable 5.2 is secretly being tested in cloud code. Now, which is interesting to hear because we kind of thought that Anthropic was trying to paste the frontier of AI development, but I guess the competitive pressure from OpenAI with GPT-6 Astra might be a little bit too intense for them and that is why we’re seeing them gearing up to drop new models. We actually have some idea on when these new [music] models will be coming as well, which I’ll talk about, but today I’m going to show you guys a couple of outputs that the new models actually two, Fable 5.2 and Opus

00:32 5.2 are producing because both of these models are secretly being tested. How are people realizing that it’s secretly being tested? Because of a simple test, which is asking the model a question without using the web and, you know, without using any memory to answer a question that’s past its knowledge cutoff date. The particular question that people have been asking is who exactly is Tibo, the reset guy, don’t use any memory and web. And Tibo, if you guys are not aware, is a employee at OpenAI who’s the lead

01:03 engineer who handles the Codex team. He’s also very vocal on X and he’s also the guy who also helps us all by resetting our limits on Codex. And we can clearly see that the Fable 5.1 model, which shouldn’t be able to answer this, is able to answer this question. A lot of people have been testing this and if your model is able to answer this particular question, then you know it’s secretly being routed to Fable 5.2. If not, then you probably are not being routed to the newest model. So, run this

01:34 test yourself. If you get a yes from this answer, maybe generate something and then see the output. A lot of people have been noticing that the output from Fable 5.1 versus if it answers this question, technically, it means Fable 5.2 is significantly better compared to what you would expect. And that is why a lot of people have been sharing their demos that they have made on X and I’m going to show you guys a couple of examples. Hey guys, before we continue, I just want to thank you guys for all the support. We have officially hit 31,000 subscribers on YouTube. You guys

02:06 subscribing to the channel means a lot to me. It helps me continue making these videos. So, if you haven’t already, please subscribe to show some support. Thank you guys once again. Now, let’s get back into the video. So, the first thing I talked about is how you know that your model is secretly being routed. This is a simple test. If your model knows Tebo, which is kind of funny that Tebo, the guy from Codex, is kind of being used to evaluate this model, then you know that the output you’re going to expect from the model is going to be stronger. One particular example is the one that you see on

02:36 screen. Fable 5.1 that doesn’t know Tebo is generating the PS5 SVG controller on the left, and that does not look anything like the one that looks like on the right, which is the one that knows Tebo, which is probably Fable 5.2. The one on the right actually looks like pretty much the PS5 controller. The only thing it’s missing is probably the logo that you see on screen. The but other than that, this model is actually generating the controller the way it’s supposed to, which is kind of funny because Fable 5.1 is getting that logo pretty nailed down,

03:09 and then over here we don’t see that logo. But, that’s fine. I’m going to cut it some slack because this actually looks like the controller versus this kind of looks like one of the old Nintendo controllers, if you guys know what I mean. Maybe some of you guys watching this might not know, but yeah, I’ve played with these in the past, and yeah, that looks more like that versus this. But, that’s one example. And we also have a example of this being tested against GPT Astra Max. So, this is the famous pelican riding the bike test. So, Opus 5.2, which is also a model that’s

03:40 being secretly routed in Claude code when you select Opus 5, you can see that this generation is much stronger than what Opus 5 is able to create. When you compare that to GPT Astra Max, Do see it definitely has a different design taste compared to the Anthropic model lineup because we can see like it’s not adding any backgrounds or things like that. But, the Anthropic model lineup, they’re generating backgrounds. They’re adding, I guess, like a boat, a lighthouse because we see that in the Fable 5.1 output as well. And you can see that To be honest, this

04:11 test is kind of getting hard to really determine if the model has gotten an upgrade because most models nowadays have gotten really good at this test. So, I think in my opinion, it’s time to retire this test, but I just wanted to show you guys that when you’re writing to Fable 5.2 with the Fable 5.1, your output definitely it looks a little bit better. But, what in my opinion, this is kind of funny, the Opus 5.2 is my favorite out of all of these. Obviously, GPT Astra Max looks like a illustrated version of the pelican riding the bike, but this one I think is probably the

04:42 best one in my opinion. I don’t know about you guys. And here’s another example where both Fable 5.2 and Opus 5.2 are kind of tested against each other. And they’re running the same rocket test. And what I can tell for sure is that the outputs look way more realistic compared to before. I think Fable 5.2 is definitely better in this output versus Opus 5.2 cuz Fable 5.2 kind of looks like something you would expect from a game. So, the area that we have seen both of these models excelling is three generations, and that

05:13 is clearly evident with these rocket tests. Because the output, as I said, look realistic, look like something that, you know, you would expect in a game or something like that. And not just a typical game, like a high-end game. Obviously, it’s not perfect, and we don’t expect these models to nail that down yet. But, the rate of improvement that we have seen compared to the previous model is clearly evident because these are way stronger outputs. Here’s the famous Vauxhall Garden test. And once again, Fable 5.2, when it comes to realism, elements, design, it looks way stronger compared to even GPT 6

05:45 Astra. But, one thing to understand is that are just eye tests. We don’t know how efficient these models are because one area that Anthropic’s model kind of really struggle with is efficiency. They tend to be on the more expensive side, and you’re paying more per, like, I guess, intelligence per dollar versus GPT-6 Astra. You’re probably getting more for your dollar. That’s just me guessing at the moment. We don’t know what the pricing is going to be for Fable 5.2 or how efficient this model is going to be. But, one thing that’s really clear is that it’s definitely

06:16 better than the current Fable model. The output Some people are saying look better than Astra. I think they’re pretty much on par. It is already people are saying slower and expensive. So, yeah, I don’t think that area is something that you’re going to expect Fable 5.2 get better at. It’s still going to be slower and more expensive than the other model lineups. But, hey, if you’re into Anthropic systems and you’re using Fable models, if you got the budget for that, you’re going to be happy with Fable 5.2. And then, here’s a game that I really was kind of shocked about because when

06:46 As I said, 3D animation is the area where all of these models are focusing on, and Fable 5.2, this looks like a pretty, like, complete game. Like, if this was a mobile game, I think this would look like something that somebody, you know, spent like a couple of months creating and everything like that cuz the way everything is set up, it looks pretty real. It looks pretty, like, production-ready, in my opinion. And obviously, these are just quick tests, but you got to give these models some credit because one-shot generations for things like this was something that was

07:17 not possible literally like maybe 3-4 months ago. But, now we are seeing that these models are able to do this quite rapidly. So, yeah, as I said, “Pacing the frontier of AI.” They say that they want to do that, but clearly not because these models are definitely stronger than anything that we have seen so far. Obviously, these are not like the model model. That’s not going to be released yet with Fable 5.2, which is going to be probably based off of Fable 5.1, just probably more

07:48 training, post training, and things like that. And that’s probably the updates that we’re going to get with this model. Cuz obviously it’s not going to be a new base. I don’t think we should expect that from them. And as I said, it’s not only the Fable model lineup, we’re also seeing Opus getting an update. And both of these models are really exceptional at 3D tasks, particularly in environments like Roblox Studio, which is interesting. And this is like a 3D generation of a world. Uh I I don’t really understand what the world is. Looks like a pool, things like that, but yeah, it’s a pool. Uh pretty

08:18 good output. Not going to lie, it’s like pretty solid and like everything looks pretty coherent. So, got to give it to them. Now, as I said, we have some idea about when this model is going to come out because if we look at this article, Anthropic was planning on IPO’ing sometime in September. Because of GPT-6 Astra and the pressure that OpenAI basically put on Anthropic, they’re planning on delaying their IPO to November, which is going to be interesting because that kind of gives us an idea that either these models, Fable 5.2 or Opus 5.2, could be

08:52 releasing closer to the November timeline, not probably soon. But that’s just me speculating, you know, because they probably want to drop this closer to their IPO, build up the hype, and that will allow them to obviously IPO at a strong price. Versus if they are already being defeated by GPT-6 Astra, and if they continue with the current timeline of IPO’ing right now, you are going to get some disappointing results from the IPO price. So, I think that is why they might be delaying their IPO to November, and that tells you how

09:23 competitive this space is because you can literally be ready to IPO, but then your competition, OpenAI, drops a really strong model, and that basically changes your whole plan. So, these models’ success are really important for these labs, which is like a no-duh moment, but now they have a really strong incentive to make sure that they drop a really competitive and strong model because of their timeline for IPO-ing. And what’s interesting is that OpenAI is not slowing down. Tibo, the Codex guy that we talked about, mentioned that we have

09:53 enough content for maybe three dev days. So, their dev day is coming up very soon, and we kind of seen Tibo mention it, even Sam Altman talk about big things are coming. So, this dev day for OpenAI is going to be significant as well, not just for that company, but also for Anthropic because this will determine if the IPO even gets delayed to maybe next year. Because obviously OpenAI and Anthropic very competitive, both trying to IPO at probably soon. We know OpenAI was planning next year, but Anthropic was planning this year. So, this dev day is going to tell us a lot

10:24 because it’s going to tell us how competitive OpenAI has been, and who’s going to set the tone for the rest of this year because we’re already in the month of September. It’s Q4, and both of these companies are pretty locked in at the moment. Even Google DeepMind is planning on dropping the Gemini 4 model. So, it’s going to be a really interesting and competitive Q4. But, that’s it for today’s video. Make 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

10:55 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.