Claude Opus 5.5 Is Insane… But Muse is EVEN Bigger transcript

You might have heard the news. OpenAI and Anthropic have released new models that are blowing people away. Some are saying Opus 5.5 is better than Fable despite it being cheaper and faster. Today, we're going to talk about both of these models, what they can do, and how you can use them most effectively. But we have more updates to talk about as well in the world of AI agents, like OpenAI's Realtime Voice being able to use all of your tools.

And maybe biggest of all, Muse by Meta becoming the fastest growing platform in the world. And Zuck might be on the verge of the next chat GPT moment. I have so many thoughts about all these updates. I've spent $3,000 in tokens over the past 2 weeks trying to test these new models, implementing these new tools in my business. So, it's time for a weekly update in the world of AI agents. The first big update in the world of agents are the releases of Anthropic and OpenAI's new models.

In the Claw desktop app, you will see their new model, Opus 5.5. And in the chat GPT desktop app, you will see their new models. They have two new models. And you can select here and you will see GPT6 Soul and GPT6 Luna. And so all of this was yesterday. Enthropic released Opus 5.5. And on the same day, two hours later, OpenAI released GBT6 Soul and Luna. And this kind of reminds me of about 3 weeks ago, when 3 days apart from each other, Enthropic released Fable 5.1 and OpenAI released GPT6 Astra.

So, we can think of that as round one. And since GPT6 Astra at the time was the best model I've ever used by far, that round went to OpenAI. However, the best model released by a wide margin yesterday was Opus 5.5. And so, both of these model releases will allow you to save more tokens and work faster compared to using these two models right here, Fable 5.1 and GBT6 Astra. But since Opus 5.5 was significantly better than previous versions of Opus, I want to spend most of my time talking about why this model is so much better.

So, the first reason that Opus 5.5 is so much better than the other models is the writing. So, Anthropic put a ton of their effort in this model release on its writing style. And so Claude announced Opus 5.5 communicates more naturally addressing some of the most common feedback we heard on Opus 5. And this is not something that you'll be able to just like see in a presentation. You actually have to feel how powerful this is.

And this is something that I've realized as I was using Opus 5.5 is when you ask it a question and there's a long chunk of text it needs to give you, it always leads with the most important part and it just feels like everything is more concise and significantly easier to talk to, especially about complex topics. And I think this is why we've been seeing this meme circulate around Twitter because everyone loved Opus 4.6.

In fact, that was the biggest opus jump ever until yesterday. 7 months ago, this model was actually genuinely fun to talk to. It would speak to you more naturally. And since then, like every few months, they've released a new version of Opus, and it has felt dumber and dumber and dumber until the latest release, which was Opus 5.5. It feels like the smartest model I've ever talked to. It's just concise and gives me the answer.

However, it's not just writing. The biggest coolest thing that I've seen that this model is significantly better at is creating videos purely in code. And Opus 5.5 created this video. And this video is a single JavaScript file. Check this out. >> [music] [music] >> And so I had to try this for myself with Opus 5.5. And this is kind of a template that you could use. If you want to pause the video, I'll actually put this prompt down in the description below this video.

You can create these animated videos where the entire thing is just a JavaScript file. And so I went back and forth with Claude Opus 5.5 and it just generated,300 lines of code. As you can see here, it just created this do.html file and then it created another file that rendered the video and the final video looked like this. Anyway, my video was not nearly as good as this guy. I highly recommend following this Kevin NGO guy on X.

All of his videos are insane. I mean, look at this. Anyway, [music] I just think these videos are awesome. And then the third thing that I noticed using 5.5, and this is the one that I use the most, is it's just so much better at front-end design and creating artifacts. Specifically, any artifact that you want to create that's very visual. It's just an incredible visual model. As a very quick example, I'll show you what I created in the new Claude projects feature.

It basically allows you to chat with a main agent. Here's Fable 5.1. and then it spins off these chats. And also they integrated claude design into this new claude projects feature incredibly well. So here it redesigned my app and instead of giving me a link with the design that I wanted, it just sent me a clawed design project. We just had a ton of different iterations on this and I was going back and forth with Opus 5.5 until we ultimately landed on this design.

It designed the mobile view as well. And this is just like a teaser page for my site, but it is so much more fun and faster to iterate with the model when you're designing things. And here's a slide deck that it generated that I hadn't checked until right before this. I just asked it to create a slide deck from my phone and this is what it created. And I just think the designs are just are just so much better in especially in slide decks in documents.

It's really fun to read. And I don't think we've even scratched the surface on how you could truly test how good it is visually. In the comments, please let me know like what things you're creating with this model. I genuinely want to see what you're building with it. We cover a lot of new agent tools on this channel, but when you actually sit down to use them, you still have to answer a pretty basic question. What work should I hand off to them?

HubSpot put together a free playbook called AI Agents Unleashed to help answer exactly that. It shows you what to hand off, what still needs your judgment, and where agents can actually save you time. Inside, you'll find practical ways to use agents across content, marketing, sales, and operations. things like taking a video transcript and turning it into posts for different platforms, researching a lead before a sales call, or organizing customer feedback so you can see what people keep asking for.

Then it breaks down to how to get started. Choose one task, give the agent the information it needs, test the result, and decide where you still want to review the work. That human review part matters. If you're spending more time fixing the output than doing the task yourself, you need to change the workflow. Grab HubSpot's free AI agent unleash playbook through the link below and put these workflows into practice. And thanks to HubSpot for sponsoring this video.

But we have a lot to cover today and we have to move on because OpenAI released a new feature to their GPT voice from the app and the computer. Your realtime voice model can now use plugins like email, calendar, and Slack. So you can use their best models, GPD6, Astra, Soul, and Luna, the models they released yesterday, and they can actually use all of your plugins, meaning all of the apps that you can use in work. Your voice agent can also use those same tools.

Let's look at an example. So, if you're in your iOS app, you can use GBT work and you can actually connect any tool that you can possibly think of to your chat GBT. Many of you have already done this. I'm using plugins. As you can see right here, look at all of these plugins that I have. Well, if we go back to chat GPT, I can press this real-time voice button. Hey chat GBT, can you give me an update on the things that have just recently been said in Slack? >> Yeah, I'll check right now. >> And look at this.

It shows you what it's doing. So, it says searching recent Slack messages. >> All right, thanks for waiting. Here's the latest from Slack. New IC is 5:37 p.m. today. >> Hey, uh, Chat GPT, I need you to do something pretty urgently. I need you to send all of these to Emily in Slack via DM. Can you do that? >> Okay, checking on it. Putting that together. >> None. >> Wow. Thank you. So, you sent that uh to Emily? >> Yep.

I've sent Emily a DM with today's key updates and links to each message. >> Hey, GPT. Uh you know the document that we were working on? It's called weekly update uh Friday morning 8 a.m. You you you'll be able to find it in notion. Let me know when you see it. It's uh it talks about opus 5.5 and GPT6 soul etc. >> Let me check. >> Look at that. You can see the tool calls. This is really cool. >> Yep, I've got it open.

Uh what do you >> Yeah. Can you please add some details under the second big update in the world of agents where I talk about GBT voice? Just give me I you know I I didn't put enough put a little table below that blue heading and and just give me some updates about it. >> Sure. Checking the latest official details now. Just a sec. >> You can see it's searching the web. You see that? It just added it right here. Look at that.

It added the table. So, you added it. >> I Yeah, it's in. I added a five row table right below the blue GPT voice heading. Connected tools. >> Yeah. Actually, can you please put them into tabs and then just have like three or four words in the actual like tab title for each one? And yeah, have all the info in tabs instead of the call out. >> Got it. I'll swap those call outs for tabs. >> Just the updates t call out, by the way. >> Right.

Just the updates one. Okay. >> Oh, nice. You did it. Look. >> Okay. Looks like it's done. The update section is now a six tab block with short titles. Everything, including the voice table, is tucked into its matching tab, and the other callouts are untouched. Very cool. Thank you. >> Anytime. >> I'm not going to lie, that's insanely seamless how easy it is that this model just uses tool calls and you can talk to it while it's using tools.

Now, it's time for what I would say is the biggest update of the week, and that is the absolute takeover of Meta's new personal agent app, Muse. So Muse is Meta's version of Grobbot, which is this personal agent platform that can just kind of do things for you. Not only can you ask it to do things and it can handle your email, control your notion, it can use a browser, have access to a computer, but both Muse and Grockbot will do things proactively.

However, Muse has seemed to do it significantly better because it is number one on the app store. If you look at it right here, Muse has passed ChatGpt as the top downloaded app on the app store. In order to find Grockbot, we actually need to scroll all the way down to the bottom to the 196 spot. And that's where Grockbot is. So Muse is a web app, a desktop app, and an iOS app where you can chat with your own little Muse agent.

And this agent has access to a computer. And very similar to chat GBT work or claude co-work, you can ask it to do anything. It is connected to all of your existing tools and it can just kind of do everything for you. And the reason why people like this app so much is it is incredibly fast. The model that they use, which is Muse Spark 1.3, Meta's model, which I think will become open weight, is very fast. And then the app is very fun to use and it's very proactive. like it'll literally give you ideas.

It'll also keep track of your goals and it just kind of does things proactively and it has a little archive or a library with all of the assets that you create inside Muse. So, this app is very fun to use and it is free. You get a 100 million tokens a month. So, I have not paid a single dollar to Meta and I've been using it a lot inside my business. So, that is why people are using it. It is literally free and I don't know how other people are going to compete with this.

So, I'm going to do a more in-depth video talking about like how to use it, but I want to talk about some breaking news about Muse and Meta, which is their Meta Connect, which just finished an hour ago. Every year, Mark Zuckerberg goes on stage in California to announce all the updates that Meta has released. And this conference was all about Muse. In fact, it wouldn't surprise me if Meta changed the name of his company from Meta to Muse.

And Mark Zuckerberg said this. Here's everything I announced at Meta Connect today. So Muse, your personal agent, now has voice and realtime video. You can have long conversations with Muse. And Muse will work in the background and get things done for you as you're talking. We're also bringing Muse to glasses so you can talk handsfree throughout the day. I don't think people realize how big this is. So, a few years ago, I first tried the Meta Ray-B bands and they were a little clunky.

You could only take videos from your glasses for about 1 minute before they would automatically turn off. So, there's a camera. There's also a microphone. So, you could like you could just say, "Hey, Meta." And then you could ask it a question. And because there's a camera, it could give you contextual responses. So I could look at a document and be like, "Hey, like what is the meaning of this?" Or like, "Hey, can you search more information into this?" And it would just see what you see.

And it could use that information when doing web search. However, AI has always been bad in the meta-bands. Like meta has never had a good AI model until recently. Now, these Meta Ray-B bands will be connected to Muse. Straight up. Meta just released may be the best consumer agent platform ever released. And now it is going to be inserted directly inside the Meta Ray-B bands. And you might be thinking like, okay, that's pretty cool, but I just, you know, it's creepy to have a camera directly on your glasses.

I'm never going to use that. That's fine. announced two hours ago. They are now shipping the first pair of audio glasses. Look at how small these are. There's no camera. Our lightest frames yet with no camera and all day battery. This is what Mark Zuckerberg has wanted his entire career. If you're a user of any of Meta's products, in order to interact with them, you have to first download the app. And most people in the US have iPhones.

So, in order to use their apps, you have to go through the app store. And Mark hates Apple. He's wanted to get out of the shackle of being on their devices for so long. And so, that's why he's going so hard into the glasses. But it actually doesn't stop there. Mark Zuckerberg just announced a little handheld Tamagotchi style pendant or I'll let you see it, but it's completely powered by Muse. We packed the whole Muse experience, including the whole real time voice and avatar stack into something that fits on its keychain. >> Oh my god. >> And is always available to talk to. >> You just go ahead and tap here in the fingerprint sensor in the corner and you can start talking >> without having to unlock a phone or open an app.

They not only are they releasing the glasses where you can talk to Muse directly and it can see what you see if you have the the Meta Ray-B bands with the video camera or it can just hear you if you have the audio only metaray bands. But they're also releasing this pendant. So we packed a lot of technology into this little guy. So if you're not wearing glasses, this is going to be by far the fastest way to talk to your muse and to show it what's going on around you.

Um, we've actually only built a few of these so far. It's It's been absolutely joyful to play with so far. Um, but we are planning to have this ready to ship uh in time for the holidays in December. >> So, they are not even done building out this product and they are racing the company to get this product out by Christmas this year. You will be able to buy one of these keychains. I'm telling you, this is one of the biggest, if not the biggest opportunity in the history of the company at Meta, because this is why he was investing hundreds of millions of dollars into Nat Friedman.

There's another guy that they had spent $250 million on, just a guy. It's why they acquired Scale AI and Alexander Wang for like $15 billion. They wanted to build the app that was your portal into personal super intelligence. And then an Nandanda he posted bruh muse is literally openclaw for normies it's actually just openclaw and so meta actually took so much inspiration from openclaw at the beginning of this year and they tailored their work that they were doing on muse to be very similar to openclaw.

Here's what Nat Freiedman had to say. Remember they spent hundreds of millions of dollars on this guy. So, we built Muse from scratch, but it's definitely heavily inspired as a product by OpenClaw. After I used OpenClaw in January, I bought hundreds of Mac minis for the MSL team and lots of us fell in love with using OpenClaw and other personal agents. And Peter Steinberger is a genius and his harness was pioneering from the jump.

I think a lot of people were inspired by it. Our goal with Muse was to build something like OpenClaw that could make it safe and secure and easy to use and scale to billions of people. So Muse is basically taking all of the good things about OpenClaw. Remember all the people were buying Mac minis and they're trying to make it accessible for billions of people. And now they're leveraging Facebook, Meta, WhatsApp, all of these platforms to get people on using this new agent.

And with this new agent, they're trying to get people to use other hardware not created by Apple to interact with Muse there. And then last week, they released their connector platform, which you can think of as an agent app store. And they already have 2,000 submissions. And I would bet you by the time 2026 is over, every major company you would ever think of are going to have a little agent app on Muse. So, you'll be able to control any app you use right now.

And I'm telling you, this is at the center of the meta company. So, if you passively watch my videos, I highly recommend just trying Muse because it's free. It's really easy to use and get set up and it is a growing ecosystem. There's going to be more apps that you can connect to your Muse very soon. This is a great place to start if you want to get used to using agents for personal work or light work use. Their models aren't quite as good as OpenAI's models and Anthropics models.

And in fact, they're actually planning to release a competitor for Muse, which we'll get to next. And for the fourth and final update in the world of AI agents, we now have it on pretty good authority that OpenAI, Anthropic, and Cognition, the company that created Devon, are releasing their personal agents within the next month. There are people actually posting on X. There are rumors of this new project either named or codenamed AON spelled AEO N getting leaked around open AAI.

And this is a new Grockbot and Muse competitor. Apparently, it's more than just a chatbot very similar to Muse and Grobbot. It's going to have its own pro profile, DMs, and groups. It's designed to compete directly with Grockbot and Muse. It could launch as early as Thursday. It also could launch at OpenAI's dev day. And apparently they're building an entire social layer around its AI. And this could also be another reason why Meta is so aggressively pushing Muse right now is they do not want this platform to catch on.

And I'm not going to make any more speculations beyond this. I think Signal kind of said it very well here. He said every major company is arriving at roughly the same offering. Persistent memory, email, calendar and messages, browser and computer use, background tasks, proactive notifications, voice, app and tool execution, meaning you can connect it to all of your existing tools you already use. Ambient context, which is why glasses are so important, which is why voice is so important.

If you have objects that are listening to you, it can allow the agent to give you more relevant notifications and message you at the exact right time. And all of this is kind of wrapped in a personal agent. And so all of these companies, Meta, SpaceX, Google will probably do something like this. Open AAI, Cognition, Enthropic, all of these companies are they're kind of arriving at this form factor which is like a safe and easy to use OpenClaw.

And I couldn't be more excited to continue covering all of these platforms because I think they're all going to have to make some trade-offs very soon. Enthropic is going to have to decide, do I want to make it super powerful and allow people's personal agent to spend a ton of tokens or do do you want it to be really fast like Meta Muse? And so all of these companies are going to be competing along different accesses like what type of user are they going for?

Is it going to be used for work or only personal? It'll be very interesting. But my goal and my job is I will try and make everything as easy to understand as possible. I'll be doing tutorials on every single one of these platforms that comes out and I will try to keep you ahead of the curve to understand how to use agents effectively. Right? You should become agent native. It's never been more important to understand agents and how to use them in your life in human history.

Anyway, thank you guys for watching. We'll see you here in the next