Gemini 4 and Muse Wonder Dropped… But Fable 5.5 Is TERRIFYING

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 and Muse Wonder Dropped… But Fable 5.5 Is TERRIFYING · Channel: Riley Brown · 25:55 · Watch on YouTube

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

Summary

The video reviews a busy week in AI, including Meta’s Muse Wonder and Google’s Gemini 4 Argon, which the host considers less capable than Claude Opus 5.5 and not compelling enough to switch to. It also covers updates to Claude, Codex, and Grokbot, with particular enthusiasm for voice-driven workflows that let users delegate tasks while walking. The host discusses rumors of an upcoming Fable 5.5 release and predicts that AI-assisted game development will enable streamers and audiences to build and refine games together. Microsoft’s Xbox Labs is presented as an early sign of a more open approach to AI-powered game remixing.

Key points

Meta released Muse Wonder and Muse Wonder full, but the host says they are priced like Claude Opus 5.5 and do not match its performance.
Google released Gemini 4 Argon across the Gemini API, Anti-Gravity, and the Gemini app; the host found its generated Call of Duty-style game worse than one made with Opus.
The host recommends Claude Code or Codex for demanding work, describing Opus 5.5 as the strongest model in their experience.
Claude’s desktop app received a design update, which the host says makes Claude Code feel faster and more inviting to use.
A Fable 5.5 release is rumored to be imminent; the host expects it may cost substantially more than Opus 5.5.
Codex updates include usage resets, creating a dot from the iOS app, a larger context window for GPT-6.1 Soul, and reduced desktop app disk usage.
The host highlights talking to an AI coordinator through AirPods while walking, using it to launch multiple Codex sessions and delegate tasks.
Grokbot added email creation, native X search, a new voice model, a faster and higher-resolution virtual computer, and improved email drafts.
The host predicts that streamers and their audiences will use AI coding tools to build and iteratively improve video games together.
Microsoft’s Xbox Labs is exploring AI-driven ways for fans to remix familiar games.

Chapters

Start	Chapter	Summary
00:00	Muse Wonder: release, pricing, and access	The host introduces Meta’s Muse Wonder models, reviews Meta’s benchmarks, and explains their pricing and ways to access them.
03:39	Gemini 4 Argon and model comparisons	The host demonstrates Gemini 4 Argon in Anti-Gravity, compares its game-building results with Opus, and shares recommendations about which models and tools to use.
09:02	Claude updates and Fable 5.5 rumors	The video covers the refreshed Claude desktop design and the possibility of an imminent Fable 5.5 release.
11:50	Codex updates and voice-driven work	The host outlines Codex resets and feature changes, then describes using a voice-enabled coordinator and Codex sessions while walking.
15:41	Grokbot’s new capabilities	The host demonstrates Grokbot email, X search, voice, virtual-computer, and email-drafting updates.
21:07	AI-assisted game creation and Xbox Labs	The host discusses a community-developed battle royale, predicts collaborative AI game-making will grow, and covers Xbox Labs’ experiments with game remixing.
25:22	Closing thoughts	The host wraps up by calling the week relatively quiet despite the model launches and anticipates a larger week ahead.

Quotes

Opus 5.5 is the best model in the world right now. — 06:54

You’re not missing out on much. — 07:57

doing things with just your AirPods in is so peak work. — 14:00

streamers are going to be able to vibe code video games with their audience. — 23:49

Concepts and entities

Muse Wonder, Gemini 4 Argon, Claude Opus 5.5, Antigravity, Claude Code, Codex, Grokbot, Fable 5.5, Xbox Labs, Riley Brown

Full transcript (original language)

00:00 What an insane week we just had in the world of AI. Many people are saying that Google and Meta are now caught up and they are joining OpenAI and Anthropic on the frontier because Google just released their brand new model Gemini 4 Argon and Meta released their new model Muse Wonder. And I have lots of thoughts about these models and what this means for both Anti-Gravity and the Muse app and also how these models compare to Claude Opus 5.5 and GPT 6.1 Soul. But that’s not all that happened over the past week. Grokbot also got four new updates. Claude Code has a few new things that were added to the app and

00:31 Codex has some new updates that people are sleeping on as well. All right, we have a lot to cover today. So, the first update of this week is a brand new model from Meta called Muse Wonder. So, this was on Monday. So, Mark Zuckerberg said, “Introducing Muse Wonder, our new family of frontier models with state-of-the-art performance for personal and professional agentic work.” And there’s actually two models that were released as part of this launch and that is Muse Wonder and also Muse Wonder full. And according to their benchmarks that they

01:05 released, it is the best at personal super intelligence index. It is the best at legal work. It is the best at financial analyst research. It is the best at business workflows, long horizon coding, and ultra long horizon engineering. I will save my thoughts about this for this section right here. For these two sections, we’re just talking about facts. And so, if you took the benchmarks from Meta at face value, you would see here that Muse Wonder full

01:35 is a frontier model only behind Claude Opus 5.5. And this was really exciting because Muse Spark 1.3 was very good for how much it costs. Muse Spark 1.3 also powered the Muse app, which is currently the number one app on the App Store. If you come up here, here is my Muse agent. This Muse agent is a competitor to Grok Bot and now GPT dots and this app is really good and it’s basically free.

02:06 I’ve never once paid Meta to use this app and it does a lot of useful tasks for me. So, when I found out that they were going to release another model that is almost as good as Opus, I got super excited because that meant frontier intelligence at a cheaper price. Well, unfortunately, that’s not what happened at all. Meta Muse wonder not only isn’t nearly as good as Opus 5.5, more on that later, it is the same price at $4 per million tokens in and $20 per million

02:37 tokens out. The model is not even cheaper than Opus 5.5. And yeah, this is just really disappointing considering how cheap Muse Spark 1.3 was. I thought it would be cheaper and I honestly thought it would be better. And they also don’t have like a coding platform that you can use that has a nice UI. You need to use it in the terminal where you type Muse just like you using Claude in the terminal and it takes a little while to load and then here we can choose which model we’re using and we are using Muse wonder 1.0 and you can see that $4

03:08 and $20 per price. And so now we can use Muse wonder. And if you want to test the new Muse, the easiest way to do it is just to go to Claude code or Codex and you could say, “Hey Claude code, help me set up Muse code. I want to run it in my terminal. Please do the entire setup so I can run it.” Press enter. And then it’ll tell you that you can now go to your terminal and open up terminal on your computer and you can very easily just type Muse and then there’s going to be this like setup process where you need to sign in with your Meta account

03:39 and then you will be able to use it. And that is just update number two, where to use Muse. You can also use it in Devin. You can also use it in T3 code. And then literally, the next day, the day after Meta released this model, Google Gemini released Gemini 4 Argon. So, on Tuesday, they released Gemini 4 Argon globally in the Gemini API and it is also available in Anti-Gravity and the Gemini app. And the Ultra sub costs $100 per month, I believe. Yes, so it costs $100 per month

04:15 to get this new model inside Anti-Gravity. And for those of you who don’t know, Anti-Gravity is their kind of cursor-like tool, which is just like an AI coding tool and that’s where I used it to build this Call of Duty game. Here is the Call of Duty game that I created. This took about like 10 prompts and for those of you who watched my video on using Opus 5.1, you saw how good that Call of Duty game was. This one was worse.

04:45 >> Team deathmatch. Scramble sector results. Eliminate all hostile forces. Weapons free. Move out. First floor, ACR up. >> The enemies like are terrible. They don’t even kill me. >> Sniper in the top office window. Watch the northeast sideline. UAV recon standing by. Call it in when ready. Nice shot. ACR’s got the reach. Fuel’s down by the jet turbine. Keep that pressure on. >> Make a knife as well. Anyway, this is

05:16 the game that it created. It’s just like doesn’t It’s not quite as good as Opus and Astra. You can see that Gemini 4 Argon, which was released just a few days ago, created a much worse Call of Duty, in my opinion. And so, in order to use this model, I showed you very briefly is I would use Gemini Argon. I think you will get the most credits for this, the same way that if you use Opus inside Cursor, you’re going to run out of credits really quickly, but if you use Opus 5.5 inside

05:49 Claude, it’s really hard to hit hit your limits on the $100 to $200 per month plan, and it’s very similar on the Gemini models as well. You get way more credits using Gemini for Argon inside anti-gravity than if you try and use it on Cursor or Devin or that type of thing. You can’t do anything like create video games or any type of document or have it control your computer like you can with anti-gravity. Okay, so what are my thoughts? My thoughts are as follows. Firstly, it is their best model yet. Meta released their best model yet, Muse

06:22 Wonder. Google released Gemini for Argon, and by the way, the price of Gemini for Argon is the same, 4 and 20. There is an intro price where it is 2 and 10, right? $2 per million token in, $10 per million token out. So, basically, half the cost of all of the other models like Opus 5.5 and Muse Wonder, but after November 30th it goes up, and they are now all converging on the same price, which is $4 and $20. And my advice to you is as follows. Use

06:54 Codex, use Claude Code. In my opinion, these models don’t do anything better than Opus 5.5. Opus 5.5 is the best model in the world right now. Use Use Opus 5.5. And so, for most of my coding stuff or anything that requires a lot of intelligence, I have just been using Opus 5.5 directly inside the Claude desktop app, but I would never I would never use normal Claude, right? I would never use this version of Claude. I would only ever use Claude

07:25 Code or use another version of Claude Code that they released called Claude Code Projects, which allows you to talk to an orchestrator agent or a coordinator as they call it, and it can spin up a bunch of threads. I’m making another projects video very soon. This is how I would be doing really complex tasks using uh the best models in the world like Opus 5.5. And then here I want to explain one other thing to you. And the seventh update I need you to know and internalize is that if you feel behind because you don’t have enough

07:57 time to test Meta, Muse, Wonder, and Google for Argon, don’t. You’re not missing out on much. The only one thing that you might gain is you by talking to Gemini for Argon, you will realize that Google’s models have a completely different personality. It’s almost like autistic in this weird way. I can’t really describe it. But other than that, you’re not missing out. Opus 5.5 is better than Argon at everything. And if you come over here in just a little over a month, they’re going to be the same price. And in order to use it, you have

08:28 to get Anti-Gravity. And Anti-Gravity is $100 a month and it’s not nearly as good of an app as Claude. And I would say GPT 6.1 Soul is actually better than Gemini for Argon at almost everything. And the Codex app is way better than Anti-Gravity. So, I will tell you this. You’re not missing out if you didn’t test these models at all. They’re not that much cheaper and they’re not better at anything. So, the eighth biggest update this week is the first update on a bunch of platform updates we have with Claude, Codex, and Grokbot. And the

09:02 first update is a new updated design to the Claude desktop app. I’m starting to really like the Claude desktop app again. So, if you go to the Claude desktop app and then you go to the Claude code tab, and let’s go ahead, let’s say we’re we’re working and we go we create a new chat. Look at this cool design here. Check this out. Look at that. They have done a really good job keeping everything simple. They haven’t gone crazy adding too many features, but their app has felt kind of bland.

09:35 And now it kind of gives me that it’s just gives me this like creative energy to use Claude code, and this is just a cool design update, and I’ve just noticed this app is really fast right now. Claude desktop, especially running Claude Opus 5.5, is just the dream stack right now. I’m moving much faster, and I’m just getting way more done using this stack. So, now the ninth update in this week of using AI agents is rumors of a Fable 5.5 release either tomorrow,

10:08 which by the time you’re seeing this it might be the day I release this video, or next week. I’m not sure yet, but Fable 5.5 is coming out very soon. I’m filming this on Thursday, but uh Andrew Curran, this guy I actually really do trust this guy. What he says generally ends up being true. I thought it was going to be next Tuesday, but it seems increasingly possible Fable 5.5 might actually arrive today. I don’t know for sure if it will happen, and if they have any deployment issues it could get pushed to next week. We will find out

10:41 together in the next few hours. And so, it’s very likely that Fable 5.5 will be the same price as Fable 5.1, and so that will be about 2.5 times more expensive than Opus 5.5, but if this model is somehow even better than Opus, it would just absolutely be insane. I don’t even know how it would get better because and as I said in my other YouTube channel where I kind of just go on rants, I have stated that when I use Opus 5.5, I don’t get frustrated unless I make a clearly

11:15 stupid mistake prompting because the model just kind of glides to what it wherever I want it to go. And so, if this model somehow gets even better, I don’t even know if I’ll be able to push it to its limits because I already feel like I can’t do it with Claude Opus 5.5, but I’m excited to try it out. And I’ll probably try and make a video the day it gets released. So, keep an eye open for that. Okay, so now it’s time to get to OpenAI Codex. And update number 12 is resets. And so, the first update we have for Codex update number 12 is

11:50 that we are getting a lot more resets. And so, since OpenAI DevDay, Thibault, who basically runs the Codex product at OpenAI, surrounding the ChatGPT desktop app {slash} Codex, he was saying that every single day for 28 days since DevDay, and I believe we’re on like day 12, um there will be a new feature or update that makes the app better, or there will be a reset. So, in the past week, uh you can now create and text your dot entirely from ChatGPT’s iOS

12:22 app. For those of you who don’t know what dot is, you probably do. My dot is who I talk to every single day inside the ChatGPT desktop app and phone app, and now you can create your dot directly from the iOS app. Um then the next day there was a reset, meaning all of your limits fully reset. So, if you guys are running out of limits, you should hit this settings button, and you should see how many resets you have. You may have more resets than you think. They updated the context on GPT-6.1 Soul from 272K to 500K. Then there was another reset. The

12:56 desktop app uses about 60% less disk space. This was a complaint that a lot of people had. When you use the ChatGPT app, if you ever go, uh like and see how much memory it’s actively using, ChatGPT would just use a lot of memory. And so, they reduced that by 60%. That’s pretty useful. And so, in summary, so far they’ve released a ton of free tokens through resets and a bunch of quality of life updates. But, I’m hearing that there’s going to be a new model coming

13:26 out in these 28 days. So, stay tuned for that. Update number 13 today is there are new voices, better accents, and less checking. Often times when you’re using dot, like you say, “Hey, I need you to check my email. Tell me what it what happened. Hey, I need you to actually spin up a new codex session and make some changes to my app.” It’ll just say, “Checking.” Or like, “Looking into that.” It does that less now, and apparently it’s much more seamless and a better experience when talking to your dot and the chat GPT desktop app. They

14:00 just announced that today. And then, update number 14 is more of my own experience. And I basically think that doing things with just your AirPods in is so peak work. It is insane. I tweeted this today. I said, “My biggest obsession right now is doing work with my AirPods in. Walking and talking to a coordinator agent who can do tasks in parallel.” So, when I’m sitting at my desk, I’m way more left-brain activated, right? I am in logic mode. I’m trying to get things done. I’m efficient-driven. But when I’m walking, I’m much more

14:35 open, right? When I don’t know if you felt this, but if you walk for longer than 20 minutes, like all of your good ideas come to you. For me, it’s when I’m actually not listening to music. When I’m just like with my own thoughts, or maybe there’s light music without any words. And then, when those ideas come to me, I want to be able to take action on those ideas. And so, for the past 2 weeks now, my morning routine is to walk for an hour, and I just put my headphones in and talk to my dot. And remember, my dot has access to my computer that’s running locally on my computer. So, I can ask it

15:08 to do things in the cloud, and if my computer’s not on, that’s totally fine. But, dot’s also allows me to spin up codex sessions. And I can spin up multiple of them in sequence, right? So, I can literally just say, “Hey, I want you to spin up three Codex sessions. One that does research on the company that just reached out to me. Two, creating a little landing page that I could send them which advertises the product that I’m selling. And three, can you please update the website to like whatever you want.” You can just spin up Codex sessions. Yet, you are just talking to

15:41 dots in a single thread. And since they have the best real-time voice feature, you can just speak to an agent orchestrator or a coordinator. And it just does things for you. Put your AirPods in, walk for an hour, and try and get some tasks done. If you don’t know what to do, just ask your dot. And make sure that you have it connected to your tools so it has context over what it can do. That’s often all you need to do. And now we’re going to do a speed round of the Grok Bot updates, updates from SpaceX. Update number 15 is you can create your own email inside Grok Bot. Literally, the steps to get this set up

16:15 is say, “Hey, I want to set up a Grok Bot email. Make the email name be blank.” Now, I don’t want to sh- I’ve already done this. I don’t want to give away my email address, but it is literally this easy. And all you have to do is ask your Grok Bot to get an email. And then your Grok Bot can now send you emails from a separate email. He can also email people on your team. And this is actually really useful for autonomous tasks. One thing to mention, once it has this email address, it can actually sign up for newsletters, it can sign up for

16:47 Gmail, it can basically act as a person and sign up for things, which might be useful for you. The next update is that it connects to X. And this is a really big deal because this allows you to actually search X from Claude Code. Let me explain this. So, you used to need an X API key, which actually costs money every time you want to do a deep Twitter search. But now, as of a few days ago, Grok Bot can just search X natively. You don’t need to connect to your X account.

17:19 You don’t need to pay extra. You can just have it search continuously through Twitter. And what I’ve set up is this webhook right here. As you can see here, it’s a Claude code webhook, which allows me to run prompts like this. So, I just went to Claude code and I said, “Hey, can you use the Grokbot webhook skill to get a table of all the days of OpenAI’s resets uh slash updates, right?” Which we did in the previous section. And Claude code ran the skill Grokbot webhook and sent it to Grokbot to search Twitter. I can

17:54 say, “Hey, can you search all the latest news in the world of AI right now on X?” And it will just automatically search through X. The same way Muse gives native more permissions for it to use Instagram and Facebook, X is now allowing you to search through Twitter natively through Grokbot. And then on Agent Native, where I keep track of like the entire um landscape of AI agents and every single tool, there’s resource If you go up to resources and you come here to connect Claude code to Grokbot with webhook, you can find all that you need

18:30 to know. You can actually set this up in less than 2 minutes. If you are someone who uses Claude code and Grokbot or Codex and Grokbot, it works from any agent. So, if you use any agent like Codex or Claude code, you can now connect it to Grokbot and this will allow you to search through all of X’s posts. Update number 17, there is a new voice model coming soon to Grokbot and they actually just released it in the API. As you can see here, introducing Grok voice 3.0. I actually tested it

19:00 yesterday. There was a link to test it. I’ll put this in the description, but you can actually try this for free. And I have this sales associate, and then you can try it. >> Hi, thanks for reaching out. What are you hoping to accomplish today? >> Hey, uh I’m just uh want to know like um how can you help me out? >> Sure thing. I can answer questions about our products. >> Chill for a sec. Tell me a joke. >> Huh, [sighs] sure. Why did the sales rep bring a ladder to work?

19:34 >> Take your time. Okay. >> So anyway, I noticed this model’s like very fast. It’s not like talking to a human like OpenAI’s models feel like. It feels a lot more personal, like you’re talking to a human. This you actually feel like you’re talking to an AI, but it’s so fast. The limiting speed thing for Grok Bot will be how fast it can actually do tool calls, because it has to search your email or search your Slack or do things for you, and so that’s why the Grok Bot one might be a

20:04 little bit slower, but these voice models are getting very fast. And so like I said, the theme of today, and what I would very much encourage you, especially people who need to move around, can’t sit still, highly recommend trying to do some work with your AirPods in only. And you’re going to be able to do this much better on Grok as well. And then the final two updates are a little bit smaller. You have a basically a better computer. So the computer inside Grok Bot will be much faster. If you click on computer right here, and it’s also higher resolution. So

20:37 everything looks much clearer when you are using the computer or your agent is using the computer, it’s much bigger now. And you can just tell it’s a higher resolution, and it looks much better. Because this is like a virtual computer in the cloud. You’re basically streaming a computer right now, and this looks significantly better. And then they made some updates to their email drafts, so that if you draft an email, it actually has the correct formatting. And so if I were to fire off this prompt, asking for a draft in my email, Grok Bot will

21:07 thinking. And here you can see it shows the entire formatted card for the email so that you can very easily edit emails, you can add people on CC, and it’s just super easy to control your email directly from Grok Bot. And then, for the final update of the week, I do want to talk about the future of video games. So, Max Blade, this guy who makes content on YouTube or X, I’ve seen his content a little bit, but he has spent 3 weeks and two Claude Max accounts vibe

21:40 coding a 20-person Battle Royale game called Kong Royale. >> We’re just not ready for it. And so then Opus 55 came out and then everybody was spamming in chat. They were like, “Go back to Kong Royale. Go back to Kong Royale.” I’m like, “Exhaust.” And then finally I’m like, “Wait a second. Like people are typing in chat exactly what I need to do.” And it ran overnight and the first time through it made 60 changes and and that was when we first invented hive coding. >> So, he’s calling it hive coding where

22:11 he’s basically live streaming playing a video game that he vibe coded. And it’s like a Fortnite-style game which is called Battle Royale. I’m not a huge gamer, so don’t quote me on this. But, he basically vibe coded the game and all the people in the chat are telling him like what changes they want to make in the game. And then he let them run overnight. He basically took all the best messages from chat, let it run overnight, and then now the updates are done. And now they’re playing. >> Look at the trees, guys. You kidding me?

22:44 Oh, the helicopter feels so good on the control. This is insane. The boat feels so good to drive. It’s insane. Dude, the two-tuck radio station? Just like GTA Thursday, October 15th, 1:00 p.m. Eastern. Max Blade is having the world’s first premiere of a non-jank battle royale, minimal jank. As ladies and gentlemen, we are Anyway, I predict that like AI is going to get so good at vibe coding video games, and it’s going to get so

23:18 and it’s going to get much faster. It’s going to get much faster that streamers, right? You’re basically going to have like a streamer and the streamer is going to have all these different fans. And basically, they’re all going to be playing these same games together. And basically, all of the people who are playing the game are going to be able to add prompts that they or suggestions that they want to edit that video game. And it’s basically going to be like this this experience where these big

23:49 streamers are going to be able to vibe code video games with their audience. And maybe every two games you play and then everyone like types in their suggestions and then an hour later, you can play a game that’s much more refined. And earlier today, I showed you that I created a Call of Duty game. That game’s multiplayer. I actually played it on a live stream. Now, there was only 19 other people playing, but it was in a big map and we were playing Call of Duty that I vibe coded. And every three prompts, we were basically adding new

24:20 things to the map, adding new guns, that type of thing. It was kind of a slow process, but I’m predict that this process will get much faster and people are just going to like vibe code a game. In between matches, you’ll just make changes to the game and you’ll just iterate until the game is perfect. And so, I think we’re we’re in store for a very fun future when it comes to video games. And the final update comes from Microsoft. Specifically, the CEO of Xbox said, “Today, we’re standing up Xbox Labs.” And basically, they’ve noticed that people are using AI to decompile

24:52 popular video games and recreate video games and run them on whatever platform they want, they don’t look at this as a super negative thing. They look at this as an opportunity to like give fans the ability to remix Halo or all of their old apps cuz Xbox has a lot of signature titles that have a lot of nostalgia. They see it as an opportunity for people to like create their own versions of it. Almost like Halo Forge mode back in the day except with AI. But Xbox is taking a

25:22 kind of a pro AI stance and they are just investing in a group that will try and create some experiments around using AI to recreate video games. And so I think this is going to be a huge trend. People are going to basically be able to create whatever they want and play whatever they want. Anyway guys, that’s an update. I know there was no real headliner to this week. It was kind of a tame week despite two Frontier model releases that I don’t think you need to use. I do think next week is going to be significantly bigger. I think we’re

25:55 almost guaranteed to get a Fable 5.5 release. And by the way, if you go to the updates page on agentnative.ink and you click on the current week, you can see every single thing that I mentioned here and all of the links as well. And yeah, thank you guys so much for watching. I’ll see you here in the next one.