Gemini 4 Carbon, Google’s Next Model? Fable 5.5 Next Week

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: Gemini 4 Carbon, Google’s Next Model? Fable 5.5 Next Week · Channel: Universe of AI · 10:52 · Watch on YouTube

Topics: AI agents, AI developer tools, AI content generation, AI models, AI safety

Summary

The video discusses reports that Google employees are testing an internal Gemini model called Carbon, while Gemini 4 Argan may become publicly available next week. The host speculates that Carbon could be a more efficient or safeguarded version of Argan, but emphasizes that its identity and release plans are uncertain. The video also suggests Anthropic may release Fable 5.5 next week, citing impressive game-like outputs that the host believes may come from a newer model being tested behind the scenes. It ends with a brief mention of an OpenAI coding feature called composer prediction.

Key points

A Business Insider report is cited as saying Google employees have been trying an internal Gemini model named Carbon.
The host says an employee described Carbon as feeling like Opus 5.5 and matching its coding ability.
The host speculates that Carbon could be a distilled or more efficient version of Gemini 4 Argan, possibly with safeguards; this is not confirmed.
Gemini 4 Argan is expected by the host to become publicly available next week, based on comments and past release patterns.
The host says Anthropic may be preparing to release Fable 5.5 next week and expects it to cost less than Fable 5.1.
A Minecraft-like game demo and a detailed box-world output are presented as possible evidence of a stronger model being tested behind the scenes.
The video notes that OpenAI’s composer prediction feature aims to guess what users will type next while coding.
The host repeatedly frames release timing and model identities as reports, expectations, or speculation rather than confirmed announcements.

Chapters

Start	Chapter	Summary
00:00	Reports of Google’s Gemini Carbon	The host introduces reports of an internal Gemini model called Carbon and discusses how it might relate to Argan.
02:05	Possible Carbon release plans and agent security	The host considers whether Google might release Argan or a distilled Carbon version, then presents a sponsorship segment about securing AI agents.
05:11	Expected releases from Google and Anthropic	The host estimates when Argan might launch and predicts that Anthropic could release Fable 5.5 the following week.
07:45	Model demos and OpenAI’s coding update	The host shows game-like outputs attributed to Fable testing, reiterates expectations for upcoming releases, and describes OpenAI’s composer prediction feature.

Quotes

an internal employee has been trying a new version of Gemini 4 which is actually named Carbon. — 00:32

This model is not publicly available. — 01:02

These are early news about it — 03:38

it looks like Gemini 4 argan should be publicly available most likely next week. — 05:42

Basically, it’s trying to guess what you’re going to type before you actually type it out. — 10:21

Concepts and entities

Gemini 4 Argan, Gemini 4 Carbon, Fable 5.5, Opus 5.5, Arcade, Google DeepMind, Anthropic, Codex, composer prediction

Full transcript (original language)

00:00 We have some new news today. Employees inside Google have been trying a new version of Gemini 4 internally named Carbon. Not only that, it looks like Fable 5.5 might be dropping next week. So, let’s get into it. So, Google has not dropped their newest model yet, which is Gemini for Argan. And the only thing we know about this model is that we keep hearing tweets on X about how it’s coming soon [music] and it’s going to be publicly available for everyone pretty much most likely if the estimates are correct [music] next week. But right now what we are

00:32 hearing is that according to a article from Business Insider, an internal employee has been trying a new version of Gemini 4 which is actually named Carbon. So I guess that’s the new naming convention that Gemini is going for. But based off what this person has been saying, this model is supposed to compete like Opus 5.5. He actually describes the words like feels like Opus 5.5 and this is supposed to match its coding referring to Enthropic’s most powerful model for long-term agentic

01:02 coding because Opus 5.5 if you’re not aware is now better than the Fable 5.1 model. And that it tells you a lot about where Google Deep Mine is. Now when people heard this news, there’s two ways you can look at this. Number one, people were thinking that, you know, Google Deep Mine is trolling because we have seen many claims that they were going to drop Gemini 3.5 Pro only to wait a long time until they are ready to drop Gemini 4. This model is not publicly available. We do know that this model is, you know, existing. It’s going to be coming out

01:33 soon. Now, internally, if they’re using another model called Carbon, and based off what I have seen, this model, I guess, is supposed to be a more efficient version of Argan. Because if I understood everything correctly from the post about Gemini 4 argan, that is supposed to be the state-of-the-art model. And it looks like carbon might be a lower version or a lower tier of that model. And if we go back to chemistry class and go back to the periodic table, the number on argan, the periodic number is 18 and carbon sits at six. Hey,

02:05 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. So, it’s pretty likely that they’re going to make Argan publicly available, or they’re going to drop actually Carbon. We might be getting a new version of Argan, which is going to be, I guess, a distilled version of Argan, which is going to be

02:36 Gemini 4 carbon. There are a couple of reasons why this could happen. The first point being that Gemini 4 argan is supposed to be a very strong model. That’s what they’re basing this model out of. And if it has the cyber security capabilities that they mentioned in the blog post being pretty much number one, they can’t release that publicly to everyone obviously because of cyber security risk. And because of that, they might be distilling Gemini 4 organ and creating a new version of that model which they’re cnaming carbon. So when it’s publicly available, this model

03:07 might be still called Gemini 4 argan, but when we’re actually using it, it might be not the actual, you know, proper version of that model. This is something Enthropic has done as well. It’s not like this is anything new. They took Methos level model, they distilled it, which was a model basically of Mythos caliber, but with safeguards and everything, and gave us Fable. That’s the whole lineup over there. It’s the same model class, but just with safeguards and things like that. So, I’m guessing this is probably going to be a similar case. Now, these are my guesses as well. It could be very likely that

03:38 Carbon is probably the next version of Gemini 5 even. This is just like an internal base that they’re working on, or it could be Gemini 4.5. Who knows what the case is? But these are early news about it, but I just wanted to make sure that you guys heard about it before anyone else. Before we continue, quick word about Arcade, who are sponsoring today’s video, and it actually connects with what we’ve been talking about this whole month. Because agents got good this year. A year ago, you would hand them a task and watch you fumble around and give up halfway. Now, it plans, uses

04:10 tools, it runs for an hour without you, and it comes back having actually done the thing that you gave it, which is great, but it’s also the part nobody really prepared for. Because a smarter agent isn’t automatically a safer agent. It’s just an agent that gets a lot further before anyone notices something’s off. And the way most people wire this up is one API key that can do everything sitting right there in the code with nothing watching what is used for. That was fine when the agent could barely function on its own. It’s a different situation now that it can

04:41 chain 40 steps together while you’re asleep. Arcade is a runtime that handles that part for you. Instead of that one shared key, the agent acts as whoever is actually signed in with exactly the access that person has and nothing past it. So, the agent’s limits aren’t something you’re hoping it respects. They’re exactly the same limits as the user. If the user can’t see something, neither can the agent. You also get a record of every single action, what it did, for whoop, in what system. So, if it does something you weren’t expecting,

05:11 you’re not piecing it together after the fact. And it comes with thousands of tools already built for Gmail, Drive, Slack, Salesforce, all made for AI to call properly the first time instead of guessing and retrying. It works with any model, any framework, and is free to start. The link is in the description. Now, let’s get back into the video. And if you’re waiting on Argan, we got more, I guess, updates from Logan Kilpatrick today, saying that Argan is coming. Fear not. And you know, this was 3 hours ago, so I thought maybe they’ll drop it

05:42 today, meaning this Friday. But it looks like most likely we will be getting this model probably early next week based off of how the model releases have been. I don’t know if we’re going to get anything new on Saturday or Sunday. Usually, most labs don’t drop anything on the weekend. They usually wait for a weekday. So, it looks like Gemini 4 argan should be publicly available most likely next week. and carbon maybe couple of weeks after maybe couple of months after who knows how Google DeepMind is planning on releasing that. Now at the same time while all this is

06:14 happening it looks like Enthropic is getting ready to drop Fable 5.5. If you were not aware Fable 5.1 was being routed to a stronger model and producing insane level outputs that most people didn’t expect Fable 5.1 to create. Remember this model Fable 5.1 is not supposed to produce better outputs compared to Opus 5.5 and it was doing that. So most likely it was secretly being tested Fable 5.5 on the back end before a public release. And that’s something that we saw with all the

06:44 models that Enthropic dropped so far. Opus 5.5, Sonic 5.5, even Haiku producing insane level of outputs that you wouldn’t expect the model to actually produce. And then a couple of days after they dropped that model. So, it’s likely that Enthropic is getting ready to drop Fable 5.5 also next week. This is supposed to be an insane model and they’re going to make this cheaper. It’s not going to be cheaper as Opus 5.5, but it’s going to be cheaper than Fable 5.1. And this is something we saw Enthropic do with every single drop

07:14 recently. They dropped Opus 5.5, which is a way stronger model than Opus 5, but guess what? It’s cheaper than that. Same thing with Sonic and same thing with Haiku. Although Haiku has that tiered pricing structure which I talked about earlier this week, but it looks like Fable 5.5 is pretty much ready for deployment and should be out next week based off what we are hearing so far. And if you’re not aware, for example, this output that this person made, which was a I guess a Minecraft clone. I’m going to show you guys exactly how

07:45 strong and powerful this output is. And he actually put the claw artifact. So I’m going to pull that up. It is called Block Heaven. So, let’s go back to the game and I’m going to play this. And I’ll be honest, I had a pretty enjoyable time playing this. Like, look at the world. I didn’t expect any of the models to create anything like this. Like, look at the tree and everything. Look at all these generations. This is high quality. And it’s like working. It’s not like the game doesn’t work or it breaks up or anything like that. I can switch around my things. I can place things. I have a

08:16 health bar. I’ll just play around. But this tells you like the next level of models that we’re getting like Fable 5.5, even GPT Bell whenever that comes out is pretty much going to solve like game development cuz if game developers have access to models like this, they might be able to implement them and increase the amount of outputs that they get instantly. Okay, this cow is mad at me. Can I Okay, interesting. That is not a cow. I thought it was a cow, but whatever the

08:48 case is, now it’s chasing me. But I’m actually enjoying playing this. Oh, it’s a polar bear. Okay. Okay. So, that’s kind of interesting. And this is the water. You can see how strong this model is. And I can move around here. And it’s pretty coherent. Like the world is not glitching out. This is probably Fable 5.5 that created this, not Fable 5.1 cuz I would not expect an output like this from Fable 5.1. Here’s another output where this person is obviously getting the new checkpoint under the testing which pretty much happens once in a

09:19 while for some users and now we’re hearing rumors that that is also being pulled back. So you might not be getting routed but you can see this box world that somebody was able to create with Fable 5.1 and this looks more detailed than you would expect Fable 5.1 to be able to create nowadays. So that once again tells you that this model is secretly being routed at the back end and it’s most likely going to be released next week if we’re starting to hear that it’s being pulled back because that’s usually what happens. They test the model out, they route some users to it and once they’re satisfied with their

09:51 test, they pull it out on the back end and get ready for deployment for the upcoming week. So, next week we have some high hopes that we finally get Gemini 4 Argan publicly available for everyone to actually test out and then we can make a video on this and actually determine if that model is competitive as they make it out to be. Also, Fable 5.5 probably dropping next week and OpenAI. Well, they are doing some weird things. They’re dropping new codec updates and they have 28 days of new codec updates and today it’s dropping

10:21 something called composer prediction. Basically, it’s trying to guess what you’re going to type before you actually type it out. So, if you’re about to work with Codeex and you know you’re going to suggest an edit, Codeex has something like suggested improvements or things that is going to guess based off what you would type in in that moment. So, that’s the update today. Not crazy. I guess a good usability feature, but OpenAI, I want a new model from you guys. Hopefully, we get something soon from them as well. But that’s it for today’s video. Make sure you guys are

10:52 subscribed to the channel. Follow our new newsletter as well at universeai.behive.com as well as subscribe to the main channel World of AI and support us on X by following the Universe of AIZ as well. Until then, I’ll see you guys in the next