GPT-6.1 Astra Might Be Coming. Should Anthropic Be Worried?

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: GPT-6.1 Astra Might Be Coming. Should Anthropic Be Worried? · Channel: Universe of AI · 08:09 · Watch on YouTube

Topics: Software engineering with AI, Model inference and optimization, Markets and economy, AI models

Summary

The video discusses speculation that OpenAI may release a new model next week, possibly GPT-6.1 Astra, amid reports that Anthropic is testing Fable 5.5. The presenter argues that OpenAI may prioritize efficiency and that any imminent release might be a modest update rather than a major new model. Examples attributed to Fable 5.5 are presented as unusually capable, while its cost and availability remain unknown. The video also raises the possibility that Anthropic has reduced Opus 5.5’s performance, but the presenter says they have not personally noticed a major difference.

Key points

The video says Anthropic’s Fable 5.5 may be secretly undergoing testing, with some users potentially routed to it.
Posts from people associated with OpenAI are presented as hints that the company may release something the following week.
GPT-6.1 Astra is suggested as a possible next OpenAI model, though the video also mentions Astra light and Astra minor.
The presenter says OpenAI’s Dev Day updates focused mostly on features and usability, with GPT-6.1 Soul described as a more efficient model.
The video suggests OpenAI is emphasizing efficiency, so an upcoming Astra update may be a relatively minor or more efficient version.
Fable 5.5 examples shown include a simulation of the Seven Wonders of the World, a Waymo teleoperator with a San Francisco map, and an animation.
The presenter says Fable 5.5’s cost is unknown and questions whether Anthropic will make it more affordable.
The video says OpenAI’s next major model may be based on a new foundation code-named Bell.
The presenter notes claims that Opus 5.5’s output quality may have declined after launch, but says they have not noticed a major difference personally.

Chapters

Start	Chapter	Summary
00:00	Hints of a possible OpenAI release	The video introduces reports of Fable 5.5 testing and discusses posts interpreted as hints that OpenAI may announce something new soon.
02:02	OpenAI’s focus and possible Astra update	The presenter reviews OpenAI’s Dev Day reception and suggests a new Astra model could prioritize efficiency rather than represent a major leap.
03:34	Fable 5.5 examples and cost questions	Several impressive coding and visual examples are attributed to Fable 5.5, prompting discussion of its potential impact and unknown cost.
06:05	Expectations for OpenAI’s next model	The presenter considers possible names for an upcoming OpenAI model, cautions against expecting a major release, and mentions the code-name Bell.
07:06	Claims that Opus 5.5 was nerfed	The video discusses reports of reduced Opus 5.5 output quality and possible reasons, while the presenter says they have not noticed a major change.

Quotes

Hey now, you don’t know what we’re releasing next week. — 01:01

OpenAI is very focused at the moment on creating more efficient models. — 03:04

The cost for running Fable 5.5, we have no information on that at the moment. — 05:35

Next week, if they do drop a model, I don’t think we should set our expectations too high. — 06:36

Me personally, I haven’t noticed a major difference yet — 08:09

Concepts and entities

OpenAI, Anthropic, GPT-6.1 Astra, GPT-6.1 Soul, Fable 5.5, Fable 5.1, Opus 5.5, Codex, Bell

Full transcript (original language)

00:00 So, if you’re not aware yet, we already know that Anthropic’s next major model, Fable 5.5, is secretly being tested right now. If you use Fable 5.1, you might be being routed to a newer version of the model, Fable 5.5, [music] and getting exceptional outputs. And at the same time, people are wondering, is OpenAI planning to launch a major [music] release next week? Because they need to compete. On Dev Day, they dropped GPT-6.1 [music] Soul. But clearly, that model was not as successful as they intended, and a lot of people are not really impressed with

00:30 the new [music] model. And because of that, it seems like OpenAI is gearing towards dropping something new. For example, if you look at this post, Tibo on September 5th mentions, who is the person at Codex, “Astra was probably our biggest competitive advantage while it wasn’t generally available. Since they’ve used this model, their productivity jumped so much that we shifted some of our plans 6 months ahead, and we’ll shift them at Dev Day.” So, this post was before Dev Day, basically saying that, “Hey, Dev Day is going to be exciting, and we’re dropping something new.” Somebody replied, “I thought you had 6 months.” And then Tibo

01:01 replies on October 3rd, “Hey now, you don’t know what we’re releasing next week.” And that next week is the upcoming week, which kind of indicates that OpenAI is not done dropping something new on Dev Day. They’re probably dropping something new next week, as well. We don’t know exactly if it’s going to be a new model, but we have some indication of that. For example, we have another post, somebody that works at OpenAI, as well, saying that, “I’m actually particularly very excited about what we’re releasing next week. Blew my mind away when I saw that demo.” So, obviously, this person is at

01:31 the Codex team creative at OpenAI. Now, most people are guessing that the next model from OpenAI is going to be GPT-6.1 Astra. They updated the Soul lineup this week with GPT-6.1 Soul, and it looks like GPT-6.1 Astra is going to be the next model. And this might be a newer checkpoint, and this could be, as this person mentioned, response to Opus 5.5 and Fable 5.5. Because these two models are producing outputs that seem way more capable than GPT-6 Astra at the moment. So, there is a lot of pressure on OpenAI

02:02 at the moment, and that’s just how the space is. One week you have one lab excelling with new model releases, their models outperforming every other lab, and then next week you need another lab to kind of do the same because the competition is so high. And it makes sense for OpenAI to do that. They did not really capitalize on their dev day. Majority of their updates, and I made a full dedicated video on this, were more feature updates, more usability updates, and those are good, but, you know, on a dev day like that you were expecting something a little bit more powerful

02:33 from OpenAI, and we didn’t really get that from them. We got a more efficient version of Soul, which was the 6.1 Soul model, but other than that, there wasn’t much. And then we also have some more posts from Thibault. Basically, today he said that fortunately our future models will be much better at code deletion and simplification. Just in time. Which, obviously, we don’t know what that means. If that means that the next model, GPT-6.1 Astra, is going to have these features, but what I do see happening at the moment is that OpenAI

03:04 is very focused at the moment on creating more efficient models. Their whole pitch with the Soul, Luna, and Terra lineup was creating models that were more efficient than Astra. So, if they get another model, 6.1 Astra, I feel like it’s going to be a minor improvement from Astra, but just probably more either cheaper or more efficient than compared to the current model lineup. Which I think is something that OpenAI should excel in, but at the moment it looks like Anthropic has a lead with their newest models, Opus 5.5

03:34 and the upcoming Fable 5.5. Before we continue, we just launched the Universe of AI newsletter. If you want to stay on top of AI news without having to hunt for it, link is in the description. Don’t miss out. To give you guys an example, I’m going to show you guys couple of outputs from the Fable 5.5 model. The output that you’re seeing on screen right now is a simulation of the seven wonders of the world. And this is pretty much the best outputs I have seen. This seems way more complete compared to anything that I have seen from any other model out there right

04:04 now. And if you look at the features and even like the animations that the model is adding to the simulation, this is something that we did not see from any of the models at the moment. The only model that I see getting close to this type of output is Opus 5.5, which is still Anthropic’s model. Here’s another example. Someone built a Waymo teleoperator with San Francisco map, and you can actually play this, but you can see at the bottom, like this looks like an actual game that somebody spent months creating. You know what I mean? This is not something that you would expect the AI model to just write code

04:34 in one shot. This is something that probably would take a lot of developers couple of at least days or at least maybe one or two weeks to create something like this. And this has all the San Francisco map embedded into it, and it has different seasons, weathers, things like that. So, you can see, this model, Fable 5.5, is probably going to be a very powerful model based off the outputs that we are seeing. Here’s another example. This is a simple motion that Fable 5.5 is creating. Now, I think this is going to change up the AI space

05:05 a lot, but not only that, it’s going to change how people are working with Anthropic’s model. A lot of people use Anthropic’s model because yes, they’re intelligent, but they’re not as efficient. And if we are getting outputs like this from Fable 5.5, I think more people are going to be willing to pay up a little bit more if they’re going to be guaranteed outputs like this. But the thing is, obviously, the efficiency it matters a lot. The cost for running Fable 5.5, we have no information on that at the moment. Fable 5.1 is pretty much unusable for a lot of people

05:35 because of their cost. Is Anthropic going to address that with the new Fable 5.5 model? Because if you think about it, their latest strongest model, Opus 5.5, is actually cheaper than their last Opus 5 model, and it’s more intelligent. So, do we see Anthropic doing the same thing with the fable 5.5 model? Who knows? I hope they do because if they are able to get this model in the hands of more people, people will be able to improve not only their quality of work, but working with AI is going to feel way different than anything so far. Now, if

06:05 we go back to open AI, as I mentioned, some people are expecting the next model to be GPT-6.1 Astra. There’s also some rumors that the next model could be GPT-6 Astra or light. And as I said, open AI is focused on the efficiency, I guess like wave at the moment and that is why they probably might be dropping a more efficient version of Astra, which could be the Astra light model. The case is, I wouldn’t expect a major model from open AI at the moment. Their next major model is supposed to be based off of a new completely different base, which is code named Bell. And when they start

06:36 dropping those models, the models that actually, you know, are causing all these issues for open AI where the agent is hacking website, things like that, I feel like that’s when we’re going to get a really powerful model from open AI. Next week, if they do drop a model, I don’t think we should set our expectations too high. As this person says, this model might be similar to GPT-6.1 Soul or Astra minor, which was something that they were going to drop, but they didn’t drop during dev day. So, this could be the upcoming model be Astra minor and be GPT-6.1 Astra or

07:06 Astra light. Whatever the naming is, we’ll find out pretty soon. Now, at the same time, one output that I’m going to show you guys is something that you do not want any of the labs to be doing is that when you compare Opus 5.5 when the model was actually launched versus 10 days after, you can see that the level of output, the difference in quality is quite noticeable. So, a lot of people have been mentioning that there could be the case where Anthropic has nerfed Opus 5.5. Now, there are two reasons when they do that. Number one, it’s not efficient for them to serve this model

07:38 anymore and a lot of people, the demand for it is a little bit too high, so they have to nerf it on the back end to, you know, keep on model running. Or sometimes what they have done is that they nerfed their current model because their next model is coming and they are gearing towards, you know, making that model look a little bit better than the current Opus 5.5 model. So, could that be the case since they are working on dropping Fable 5.5? Who knows, but I want to know your guys’ opinion. If you have been using Opus 5.5 over the last couple of days, do you feel like it is nerfed? Me personally, I haven’t noticed

08:09 a major difference yet, but we would love to know [music] what you guys think in the comments. But, that’s it for today’s video. Make sure you guys are subscribed to the [music] channel. Follow our new newsletter as well at universityofai.beehive.com, as well subscribe to the main channel World of AI, and support us on X by following the universityofai.z as well. Until then, I’ll see you guys in the next video.