NEW Claude 4.6 Update

OD Opus 4. 6, our smartest model in the world for AI. It's just got better. And you can see the new release right here. So, it was just announced a few hours ago. This is Claude Opus 4. 6, the smartest ever model from Claude. It's just been upgraded once again. You can see the benchmarks here. Honestly, I don't really care about the benchmarks. I don't really trust them. I don't think they're very useful in real life. But you can see how they perform. If we have a look here, for example, you've got Opus 4. 6 6 versus 4. 5 versus Sonic, 4. 5 versus Gemini, and it crushes them all on all of these as you can see right here. So, this is going to be very interesting to see how it performs.

They've also got a new update called agent teams as you can see right here. So, you can orchestrate teams of Claude code sessions. I'm excited to try this out and you can also get it available on the API. You see here that Claude Opus 4. 6 six is available on claw. io claude developer platform all major cloud platforms and within co-work as well. So we can take a look at the blog post here. If you actually want to get started on this, you can go over to claude AI like you can see um and start using it here directly. What you do is just click on the drop down here and then you're going to switch to 4. 6 Opus 4. 6 and start using it. Now we can

actually compare these side by side. So, if we look at the older version of Claude versus the new one, I think it'll be honestly a slight difference, but who knows? So, if we go, for example, for Sonic 4. 5 versus Opus 4. 6, we've got them in two different chat windows. And then what we're going to do from here is we are going to grab a prompt for SEO from the AI Profit Boardroom. If we scroll down, we'll get the video to blog post section here. And then we're going to check out this prompt and plug this into Claude. All right. So if we go back to Claude here, we'll put keyword equals SEO training in London. And we'll do the same inside here as well. Exactly the same prompt. Opus 4. 6 versus 4. 5 Sonnet. Obviously two different models, so that's the easiest way to compare them side by side. We'll see how they perform. N says, "Waiting for Gemini to really up the game. " I would agree. I think they are beginning to fall behind. It's been a while since Gemini really pushed anything that blew my mind. Obviously, when Gemini 3 broke AI, it was pretty impressive, but that feels like a long time ago in the AI world. So, this is 4. 6 on the right hand side and then we have Sonet 4. 5 on the left hand side. These are both on extended thinking. All right. It's actually interesting the way that it's formatted looks different. You can see that Sonet 4. 5 seems to be a lot slower than Opus 4. 6. So, for example, here it's already written the content. It's expanding the content, proofreading it. whereas Sonic 4. 5 extended is taking a lot longer just to create part one of the article. So it does seem like in terms of speed they're very different like you can see and in the meantime let's check out the differences here. So

you see here it says on the official website the new claude opus 4. 6 improves on its predecessors coding skills it plans more carefully sustains agentic tasks for longer can operate more reliably in larger code bases and has better code reviews. So, it's very like code focused, right? That's what it's really focusing about. One of the biggest differences here is the larger context window. It's actually what, five times longer. I think previously it was 200k for the context window and now that's gone up to 1 million token context window, which is pretty insane. We got the legendary Martin, mate. I see you everywhere. So, shout out to you. You're doing good marketing, man. What up, bro? How you like it so far? I'm doing a test as we speak for mobile on all tools. This is literally the first time I've tested out, so I haven't seen the outputs just yet, but Claude always delivers, don't they? They're always good. So, yeah, shout out to Marson. We met at the school games in Los Angeles. Both won Q3 last year. Went to Los Angeles to meet Alex Hozi and Sam Ovens. That was pretty amazing. So, let's go back to this. What we got here? So, this is the 4. 6 output. This is sonet 4. 5.

Let's pull them up. So, this one has actually given us four five headline options, which is what we wanted, whereas this one only gave us one. So, already Opus 4. 6 is following the instructions better. Also, what I would say here is like if you look at the actual headlines, this headline is less relevant to the search intent, right? So, this bit right here, nobody searching for this keyword cares about that, right? What they care about is like how to rank, which is what the search intent is covered inside the headlines of each of these articles. Now, let's pull down um an example. See what we got here. Look at this. So, if you look at the this is set, the old version SEO training London never taught me what I'm about to show you. It just doesn't quite make sense, right? It feels kind of forced. And that sentence right there is just kind of weird. You you'd read that and be like, what you on about, mate? You know, um whereas if you read this intro, SEO training London is something thousands of people search for every single month, most of them end up wasting their time on outdated rubbish, right? I would say this article just feels a lot better written. Not just the way that it's written, which is actually better, but also the fact that it's got a lot more personality and also just like if you look at the actual language used and the way that it's used semantically, it feels more natural, right? It fits into the conversation more. So, for sure, Opus 4. 6 already feels better versus Sonet, which is great. Now, let's have a look at the rest of the announcement here. It says the model's performance is state-of-the-art and several evaluations blah blah. And then you can see here it's got knowledge work, agentic search, coding and reasoning. Right? So four different areas right here. If you have a look at the benchmarks, uh you can see 4. 6 is absolutely crushing on knowledge work. Let's take a look at aentic search. Yep, it's crushing on aentic search. And then coding Marian says, "Appreciate you. Great time in LA. You see me? I feel like you have 20 clones, mate. Keep rushing. I probably do have about 20 clones, something like that. So, if we have a look here as well, we got the coding benchmarks are much higher. Let's have a look at reasoning. Yeah, reasoning as well. Now, it says you can assemble agent teams on tasks together. Right. So on the API, Claude can use compaction to summarize its own context and perform longer running tasks without bumping against limits. Now compaction was something that you saw before previously if you were using the new update from Opus 4. 5. It would like sometimes it would go over the context window and it would compact the conversation to make it easier to carry on going without like just ending the conversation there. So it seems like it's using that plus uh higher context windows to basically create more and better with claude code. Um also says they've made substantial upgrades to claude and Excel and they're releasing PowerPoint as well under research preview. It's available on API as well. So it's not just available inside the chat but also API. You can see some little testimonials there from uh GitHub and notion. Then evaluating Opus 4. 6 right here on the benchmarks. So, looks pretty cool. Let's try some other stuff. Now, I'm actually going to try comparing the old versus the new. Um, we'll do a few more tests. I'm going to use the AI success lab to come up with some new tests and interesting ones. So, we're going to try this prompt right here and see what the difference is between the outputs from Sonet versus Opus. So, if we go to a new chat, we'll try this out. So, this is Opus 4. 6. and we'll compare it against Sonet 4. 5 and see if there's a big difference between them. Let's have a look at the APIs as well. If we go over to models. Yeah. So, we got Opus 4. 6 right here. And let's compare it against Opus 4. 5. All right. Yeah. So, you can see a big difference here. So, you got a million versus 200k context window. They're both reasoning models. This is faster than Opus 4. 5. So the latency is lower as you can see less latency same price actually that's interesting. So they are the same cost and 4. 6 can't work with files as a input modality whereas 4. 5 can also not just the input tokens but the output tokens as well for context are much higher. It's doubled. So the five 5x their context window for input tokens and then max output tokens they've doubled. Nice. What else we got here? One nation says thank you Julian Goldie. you're almost live 24/7. Your AI and SEO tips are incredibly helpful and inspiring. Happy to help. And then N says, "Can you make it create a website in one go? " We'll test that out in a second as well. Let's come back to this. So, we'll see what we got here. So, we got both of them still coding out the Space Invaders game. Let's set up another test in the background. Oh, actually, we've got this ready from Opus 4. 5. So, I mean, uh, sorry, Sonic 4. Sonic 4. 5. I'm losing track of all the numbers here. So we got this is the version from Sonet which looks pretty cool. And then we're still waiting for Opus 4. 6. So does seem to be taking a little bit longer. You see it's still coming out right there. Now let's try and create a website with them side by side. So this is the landing page. I'm going to plug this into both of these. Let's make sure we switch to 4. 6 of it. I'm going say create a and this is basically for my AI automation community AI profit boardroom right so I'm going to see if we can create a better version of this or if there's even a big difference between the two when it comes to oneshotting a website so I'm going to say okay create a beautiful modern direct response sales page for this website I'll put my brand colors make it beautiful interesting domain inducing all right, we'll take the same prompts, we'll plug it into Sonic 4. 5 as well, and then we'll see how these perform. In the meantime, we do have the new Space

Invaders game from Opus 4. 6. So, let's see how that performs. Well, this is looking good. Look, look at this, peeps. Look at this. So, you can see here, for example, if we check out the game from Sonic, it just goes straight into the game, right? So, there's no like entry point or anything like that. If you look at Opus 4. 6, six. It's actually created the title page of the game before people can start using it. It's got sound effects. Just going to view this. It feels way smoother, right? Feels way smoother when you play it. Looks pretty cool. I love the effects. I love the way it's designed. This is in one prompt as well. I mean, it looks super nice, doesn't it? How cool does that look? And it actually works, right? So, compare that versus this. Like, I can't even Yeah, I would say this is way more messy, right? If you compare them side by side, you're like, "Oh, this is clean but fun to play. " Right? So, I would say 4. 6 is absolutely crushed on it when it comes to the actual outputs right there. Let's try another test. Right? Now, I'm actually I'm going to close Sonic for now cuz I don't think we need that. Let's have a look what we got on the website. It's still coding out the website. That's okay. So, now what we're going to do is we shall go to create another game. By the way, if you want the prompts that I'm testing out right here, they're all for free inside the AI success lab, link in the comments description. We're going to test another thing here. So we'll try this one right

here. Plug that in. DMV says Opus 4. 6 memory is trash and it chews through tokens. So we're actually using this inside the chat. So we're not using tokens and also the memory is just built into the account. But yeah, I can imagine like if you use it on the API could I mean the same as Opus 4. 5, right? It could get expensive if you're using it on the API for sure. And Ian says use Claude co-work to test 4. 6. All right, let's test this out. So, we're going to go to Claude over here. We may need to update. Let's have a look. So, co-work over here. There we go. Oh, yeah. We got Opus 4. 6 built in as you can see. What would you like me to test it? Maybe we can just create something from one of those lists right here. We'll create this one. Just open up a new folder. If you're not familiar with Claude Co, got loads of training inside the AI success lab, but basically this is like a an AI agent that connects locally with your laptop. So from here, we're going to type in just a little folder to set this up in. Go from here, hit always allow. Then we'll plug this in says build skills and plugins. It's like a clawbot now. Yeah, I suppose it pretty much is, isn't it? Yeah, I can imagine this is very similar. In fact, if we go over here and we go to new task, right, and then inside the claude 4. 6 folder. I'm going to grab from my downloads the skillbox. zip. This is a skill that I have set up so that I can quickly create like images, videos, podcasts, etc. using Claude. I'm going to say install the attached skill skills. Then create a quick image of a lobster. Let's see if we can do that. All right. So, we've got two things coding out with Opus 4. 6 six inside coowork as well. And then if we go back to the chat directly, we've got the game over here. And we have the website over here. So you can see the website. It's not bad. Would I say it's as good as the original? This is the original right here. I actually think this is better, but it took a lot more prompting to be fair. This is just a oneshot website as you can see. Looks pretty nice though. Yeah. I mean, we could deploy that. No problem. Looks pretty nice. Look at that design. Wow. Okay. I hope it's 4. 6. six is already goated, I think, at this point. Looks super nice. Look at the design of that. The colors it's used, the gradients, the way the button flashes to get attention. It's pretty cool. So, if we go back to the Flappy Bird game as well, let's test this out. It's all right. Not bad at all. Not that impressive, but not bad. And then if we go inside here, it's now creating the typing game and the picture of a lobster. All right. So you can see it's beginning to install these skills. However, it is struggling to install those skills. Usually if you're using something like board code, it can install those pretty quickly. Let me try the documentation on the website. We've got the game ready to go as you can see. So we can allow that. Here we go. Yeah, it works pretty nicely. Pretty easy to set up. Yeah. Yeah, that's a nice game right there. All right. So, I mean it's very impressive. It's handled everything that we're throwing at it. It's been pretty simple and easy to set up. Kuno says, "Hi, bro. Good to see you. " All right. I've tried it inside Claude Co-work as well. Um, it doesn't seem to use the skill boss API very well. It seems to be struggling with that. Whereas I know for example, if you're using anti-gravity or you using something like for example clawed code, this would work way better. So I would say this is where it's failed is installing skills, external skills that I've already pre-loaded that work on other platforms and that sort of thing. So everything else was all good. Let's just try this one. So I'm going to try it one more time. Give it one more shot. Let's see how it goes. Ask it to build its own skills and persistent memory. So I mean it could easily do that just by but 4. 5 could do that too. But yeah, if you wanted to just create a persistent memory like you could say, okay, create a skill for SEO, something like that. Yeah. So it failed on setting up skill boss. See if it can create a skill for SEO blog writing or not. Yeah, look at this. It's totally failed on that as well. I would say claw coork is probably the biggest let down, but maybe we have to just update. So, I'm going to try and update and restart Claude now. Let's try this one more time. Yeah, look, it's not working at all. Didn't work with the SEO. So, I mean, like it seems really good apart from Claude Co-work. When I test out with Claude Co-work, pretty average to be honest with you. So, that's basically it for

Claude 4. 6. If you want to get a full guide on Claude 4. 6, we've actually created a full step-by-step guide here on how you can use it, what it means, etc. And that's inside the AI profitboarding which is my AI automation community link in the comments in description or just go to apiprofitboarding. com. Inside here you'll find an amazing supportive community of great people who are all focused on scaling, saving time and growing a business with AI, right? So you can post inside the community. We have 56 people online right now. 2,300 me inside the calendar. You can ask questions and jump on live calls whenever you want. On the map here we have people all over the world. So, if you want to meet people inside your local area, you can see that we have people all over the world. Um, and you can meet some amazing people who are doing similar things to you in your local area. DM them, jump on a Zoom call, meet in real life, etc. Inside the classroom, you can actually get our best trainings. For example, we have a full six week AI automation master class that takes you from beginner to expert with AI, plus how to create your first AI agent in under 5 minutes. If you're interested in Claudebot and Maltbot and OpenClaw, you can see we actually have full courses on how to use this stuff um with step-by-step guides right here. If you want to get all my best playbooks I personally use for my business, you can get them inside the playbook section here. So, for example, my X and Twitter automations, my automations for shorts, Instagram, how I automate AI avatar videos, and also if you miss the coaching calls, you can watch them back in this section. We also show you here how to get more clients for your agency. And additionally, if you want to rank number one with AI and SEO and rank inside other search engines, you can check that out right here. Additionally, you can learn how to grow a YouTube channel from scratch using this section here based on what's working for me. So, that's all inside the AR profit boarding. Feel free to get it. Link in the comments description. If you're wondering, okay, like other people getting results with this, too. So, it's not just me. So, for example, Steve Light is crushing it with this YouTube AI challenge. You can see here that Abdullah built his first NA10 automation which is pretty amazing. CJ learned how to install NA10 locally and Joseph built his own to-do list app. Right? So that was just within the last 24 hours and there's so many people winning and learning and growing with AI inside this community. Plus you can connect with me personally inside