NEW Grok 4.2? Sherlock Alpha is HERE!

All right. So, today what we're going to be doing is running through the latest stealth model update from Open Routter. Now, if you haven't seen this, basically Open Router have released this brand new stealth update. There's two different models here. One of them being Sherlock think alpha and Dash Alpha. Right? So, these are two models. Now, what's really interesting about these is if we have a look at these, you can see you've got these two Frontier models. The context window here is 1. 8 million context window which is pretty wild. And also you can actually use them for free. So you can use the APIs for free and get access to this right now. A lot of people saying this might be Gemini 3. 0 Pro. Some people saying it's Flash, etc. I'm not sure, but we're going to test it out and just see how it goes. And uh we'll run through and test this out, see how it performs, etc. Some people saying Gemini um three, some people are saying X, some people even saying it's Quen, right? Other people saying it's Grock. So, who knows what it is, what it does, etc. But there's only one way to find out, and that's what we're going to find out today. So, if you've never used Open Routter, essentially open router is like uh it's like an API kind of place where you can get access to everything in one place. All right. Uh, so we're going to have a look at this in a second. Some people are testing out here on the canvas. Let's just get straight into testing it, my friends. Why not? You only live once. All right. So, we're going to go to open route over here. And then we're going to click on models. And then you can see we've got two different models, right? So, we got Sherlock Alpha and Sherlock think alpha. So, let's take a look now. See what we got. Going to go back over to here. And you can see the two different models. Oh, the API just came out for uh GPT 5. 1 as well. Happy days. Cool. All right. So, let's check them out. Let's go with we'll go with Sherlock Dash alpha for now. So, it says all prompts and completions for this model are logged by the provider and maybe used to improve this model. And then if we go inside the details here, you see it was it literally I think it just came out 2 hours ago, right? 2 hours ago. How's that for speed? Let's have a look here. When did they release and announce this? Yeah, two new stealth models two hours ago, right? So, this is brand brand new. And you can see here all prompts and completions logged by the model. We got the context window right here. 1. 8 million tokens. Um, and then it cost you $0. Obviously, you got to bear in mind like this will be gone soon, so make the most of it whilst you can because once it's gone, it's gone and uh and sometimes you don't get it back. All right? So, just make sure that you do try and get this. — Just going to mute you, Greggo, cuz there's a lot of background noise there, mate. Um, and then — No, no. I'm going to just mute you there, Gregoro, cuz uh it's super loud in the background. So, let's see. We'll have a look here. I mean obviously like one of the worst ways to check what it is to actually go inside the models like if you go to dash for example here and you go to chat some people will say like okay what model are you and usually it will just say like I'm open AI or something like that so let's have a look here I'm sure dash alpha a large language model from an unknown provider is super fast when it replies to be fair and uh we'll keep going through it we'll have a look and we'll have a cheeky gander here and see what we can do now if we actually have a look at the overviews Here you can see some information about it. So for example, the total context, the max output, input price, etc. latency, blah blah. And if we go to providers, you can see where it's provided from. And obviously, it's a stealth model, so you don't really know. Then you got performance, so you can compare it against these models. Let's have a look versus Optimus Alpha. Then you can see which apps are actually using it. So root code is right up there. You got Pax, Histori, Kilo Code, Open Router, Chat Room, Client as well. activity just started getting used and then quick start you can grab an API key directly from uh open router right now if we go to compare here we can compare the models now bear in mind there's two different models think alpha and dash alpha all right so we'll compare these side by side so we got open router context length 1 to uh 1. 8 8 million tokens text and image input output text. So it can only generate text not uh images as well. And then it doesn't so dash alpha doesn't have reasoning whereas Sherlock think alpha does have reasoning. Damian said I tried the web OS using HTML prompt and it's really bad. I really hope this is not Gemini. What's a webOS prompt? That's interesting. I don't know that one. I'll be interested to know. Let's keep going through. So we'll go to the chat over

here and then we got a bunch of different options here. So we can create like interactive app using Sherlock think alpha and we can also use if we add another model here. So we can use think and dash right. So let's change the model over here X alpha plus dash. All right. So just to be clear here we've got think alpha and dash alpha. Reasoning is mandatory for this model. I think you can change the advanced settings here. So you can actually change like okay what's the preset filter enable reason etc. reasoning efforts. You can change this to minimum, low, medium, high. So, you get four different reasoning efforts. And then here as well. Kind of tempted to see what people are saying on X about it. You know, we got to check this out, and we Let's check the rumors. Okay. All right. So, let's see what we got here. So, a lot of people saying, "Yeah, Grock 4. 2 Let's see what else we got here. Tah saying it's by XAI. Yeah, I think the general consensus is that it's Grock 4. 2. Some people saying it's pretty poor, though. I like the UI of Shira or Skira. Oh, there's me. Look at that. All right, let's go and check it out then. Let's have a look here as well. I think we can compare it. So, let's have a look at Gemini for example. Gemini 2. 5 Flash Pro. Sorry, Gemini 2. 5 Pro. So, look at the context window difference, right? This is Gemini 2. 5 Pro which is 1 million context window. And look at the context window of Sherlock- Alpha 1. 8, right? Which means you can do a lot more without running out of context length with Sherlock- Alpha. Let's have a look at Claude as well. I think Claude's latest model is even a pretty low in context window. Yeah. So, a million versus 1. 8. Like, this is a huge step up. Step up, isn't it? Massive. All right, let's get back over here. We're going to share this tab. We'll run through some examples. Create a web app. What do you want me to create, peeps? see and test? Glenn is saying it's Max Mini Max M3. There's a wild card. I've not heard anyone suggest that. That's the first one. So, I respect the originality. What should we build today? I'm going to have a look over on chat GBT. Get some suggestions. See what we can do. I gotta say, by the way, chat GPT5. 1 for just general answers is driving me crazy. It's so bad. But let's we'll come on to that in a minute. Faceless YouTube. I'm just going to go over here and we'll just create a Flappy Bird game just for a joke. style game. All right. And we're using Think Alpha and Dash Alpha to test this out. So, let's see what we get back in a sec. Oh, here we go. We got some cool ideas. Make a DJ's console two channel with effect. Create a PS5 controller in HTML. Make GTA 6. Just build GTA 6. That's what I'm talking about. That would be a productive morning to be fair. Like, I mean, it's How long's it been since uh GTA 5 came out? It's crazy. All right. Well, we got two people request. Well, we got a request for DJ console with two channel effects. So, do you know what? I'll create two different options over here. All right, we'll create a new chat interactive app. Create PS5 controller in HTML. So, we'll do that test and then on another tab, what we can do is we will test out what was the one create a DJ console. GTA 6 equals AGI for sure. Hopefully GTA 6 comes out one day. I haven't played video. Video games are just not the same to me, but that's probably me being old. So, we got here the controller PS5 test and [snorts] um make a DJ console two channel of effect. Let's try that. So, we got Flappy Bird generating. We got PS5 controller in HTML. And we have a DJ console with two channel effect. Nice. So, it's generating. Generating is magic. What is the What's the prompt for the controller PS5 test? I've not heard of that before. Is that like an actual prompt? Feel I feel like it's a test that I don't know about. Create a billiard game. All right, let's do that as well. I love the fact that it's free. So, 3D game. Okay. Okay. All right. So, let's check this out. We got the Flappy Bird game back. So this is the one from dash. Oh, it's failed. So dash alpha has already failed on us on the first test. It couldn't even create flappy bird. I mean, if you can't create flappy bird, mate, what we doing here? All right, so Dash alpha totally failed there. Let's have a look at Think Alpha, which is obviously the reasoning model. Let's try this out. Go preview. Oh, it's beautiful, isn't it? It's a beautiful time to be alive. So going to open this up. A perfecto. This is pretty good, not going to lie to you. And hard to play, but I mean, I like the colors. I like

the design. The functionality works nicely, too. I can't get past the second one, but that's probably Oh, there we go. Yeah, that's pretty good. That was a great flappy bird. All right, so far, just to recap here, we go back. This Flappy Bird was pretty good. This one totally failed with Dash Alpha. Wouldn't recommend it. Pretty bad. All right, let's go on to the next one. So, we're going to share this and we've got create a PS5 controller in HTML again. Dash alpha. Oh, no. Which one's this? Let's go back. All right. So, dash alpha. Let's check this out. So, we got success. Let's preview it. Well, I suppose it kind of works. Let's have a look. So, this is a PS5 controller. You can kind of It's really nostalgic using this to be fair. So, you can mess around with the D-pad. Most of it doesn't work to be honest. I wouldn't say that's great. So, exit the preview on that. So that was dash alpha which is the non-thinking model and you can see here it looks really weird doesn't seem to work etc. Right now if we go back and we go to the thinking model Sherlock think alpha and we check this out. It totally failed as well. So we've had two failures on both tests. It doesn't really look like a PS5 controller. I would agree. I don't think it's that great. All right. But there we go. So PS5 controller test. wasn't that impressed, but I kind of want to see what um Gem and I can do when it comes to creating a PS5 controller. So, let's try 2. 5 Pro. Switch over to canvas and we'll say okay, see what we get back there and we'll come back to that in a second. All right, so just to recap on test two, create a PS5 controller in HTML. Dash did it, but it doesn't look quite right and think alpha totally failed. So, let's move on to the next one. So now we said create a two channel create a DJ console two channel with effect and it's failed again. Okay. Do you know what it's good? I always say it's good when you see this. Wait, did you literally create a video game on the fly? Yeah, Tina. Yeah, I did. Yeah. Tina's watching on the free school group and uh yeah, created a free um game on the on the fly. Right. So far really nasty. Nasty so far. Let's have a look with this one. So, this is create a DJ console. Let's preview it. What's going on over there? What is this about? It says preview and then it's not working. I think it's just totally failed, isn't it? All right. So, we got three failures there. Three for three. We just got a blank screen. Pair rank says it's horrendous. Tell me that at the start, mate. I just spent 25 minutes coding with the most atrocious APIs that I've ever used. I don't know why this is trending or why people are hyped about it. The only thing that's exciting is the um I can't believe how bad that is. Why do they release these models? So if that is the new Grock, I'm seriously concerned to be honest with you. That was really bad. Well, let's try out these new open router models in Trey. I'm going to have to install Trey and I want to do that. Not particularly. Has anyone ever told you that you look like — [sighs and gasps] — that I look like a British ninja. I mean, no, I've never been told that, but I like the idea of it. So, thank you, sir. We're all in the same boat here, which is we've all wasted time testing out tools that are really, really bad. So, um, let's have a look. So, we got the PS5 controller from Gemini. Honestly, that's not great, is it? Let's try Claude. That was the strongest espresso shot. I'll tell you that for free. Code Ninja. Wow, that's bad. Yeah, I would agree. So, let's test out um Claude Sonic 4. 5, which is supposed to be the best coding model in the world. We'll see. Only one way to find out, right, peeps? So, here we go. By the way, just whilst I'm uh doing this, you know, I was thinking about like content for my channel and I'm trying to figure out, okay, what's the best way to keep creating content with the limited time that I have and still create content that's useful for people? I'd love to know what you think in the comments. But basically, one thing I've been thinking about lately is like my avatar videos actually get more views and more reach and more engagement than my human videos, right? So, for example, here if I share my tab, right? Let me show you an example. So, if I create a video about Skywork uh let's say for example a new chat GPT update, it will get like 6. 2K, right? But typically when I do AI avatar videos, they tend to get more views than uh or all the same views versus my human content. They also tend to get more reach and they tend to get um the same amount of engagement or higher watch time. Plus this way I can create more content that covers more updates that helps more people if that makes sense. So what I'm thinking about doing on my main channel and let me know if you hate it, you hate it. But I just want to

know like some for example like you can see I created this human video right here and a lot of these human videos are not doing well like 1. 7K 2. 4K uh 64 views etc. So what I'm thinking is that we create long form with AI inside my main channel and then on the live streams that's when I'll be a human um and actually just cover stuff that you want to see with you in real life. And I think that's probably the best way to do it from what I've seen because when I create human content, it doesn't seem to get more views and everything else. So, I'll be interested to know what you think there. But, um, yeah, that was just something that was I was thinking about and I thought, right, let's bring it up whilst we've got lots of people here. So, if we go over to Claude here, we have the PS5 controller. That is the closest that we've got to an actual decent representation. Right. So, we can move the analog sticks as well. These buttons work. R2, L2. I mean, Claude has just crushed it, right? Claude is way, way better. Yeah. No, that's pretty good, isn't it? We got Daniel on stage as well. Greg too. Try out GPT1 thinking, "All right, come on then. Let's try it out. " I'm not excited to do this cuz I know how bad chat GP 5. 1 has been lately, but I'll test it out for you. Right. So, we're going to go to thinking and then canvas, right? So, we got thinking and canvas and we'll say right create a PS5 controller in HTML. Let's try that. So, it's doing something m I want to show you something that I've done with maybe I'll come on to that in a second. But I don't know if anyone else is having the same problem, but like for example, if I want chat GPT 5. 1 to type out URL, it just gives me like this weird h it literally writes in words. So for example, instead of typing um slash, it will actually just give me the word slash sl. And it's like, what are you doing here, mate? You know, why are we doing this? All right, so we want to preview this bad boy. preview the PS5 controller. It's trash, isn't it? It's total and utter trash. Yeah, it's just like I mean, it's really good to see this cuz then you get to see Okay, Claude is a goat by far, isn't it? Look at Claude's beautiful PS5 controller. And then look at this PS5 controller right from Chat GPT 5. 1. It's just a joke. Jason says, "The only last option is Google Stitch for the design. " Do you know what? That could be a good one, actually. Let's try that out. Let's try it out then. So, we're going to use Stitch for the design. Here we go. Use Gemini Pro. Appreciate you, Jason, man. You're an absolute legend, mate. Thank you. The sporter says, "Any Chinese models that can do a great job creating a controller? In the background, I'll open up Deep Seek and see if we can create the same thing. All right, so we got Stitch coding up the PS5 controller there. Then we're going to use Deep Seek [clears throat] Deep Seek uh 3. 2 as well. So, it's doing something. Do you know what? We'll try Quinn. Actually, you know what I'm tempted to do is use Ernie. Ernie just came out with a new model recently. I don't know if you know Ernie by Badu, but you can basically choose preview over here. We'll log in with Google. Let's try that. All right. And we'll try the same thing. So, we're going to use 5. 0 preview, which is a new model and we'll try that. All right. So, we're using Ernie to try this as well. Chinese model that supporters requested. And then if we have a look here, what on earth is this? The PS5 controller is an amazing test to be fair. Shout out to pair ranks for doing that. What is that? What even as like I just imagine like you're trying to work and you're trying to present to your team like okay the power of AI and you're like watch guys look wait for this guys you you've got to start using AI for your coding and then you pull out Google stitch and you say create a PS5 controller in HTML and it gives you back this in front of your whole team. How embarrassing would that be? So, I think the only thing for coding right now is Claude called uh 4. 5 Sonic. Go back to if we have a look at um yeah, Ernie and we ask for a PS5 controller, it just gives us this, right? Is that has that just generated an image or is that an image from the web? I think it's actually just created that image, but it's not very good, is it? All right, let's go back to Claude. So, Claude is still the goat. Sper says, "Ernie, must have forgotten you mentioned that one. " Yeah, Michael says, "I think it actually might be a meta model. " That's interesting. Jason says, "The PS5 controller test is a new stress test. " I would agree. I think that's it's really good actually that test. And then Patty says, "They say it could be Gemini 3 already. " Kimmy is overhyped. Also try

Kimmy K2 thinking. All right. Come on then. Kimmy K2. Who's a go? All right. So, we're on Kimmy K2 now as well. We should enable thinking to make it a fair test. Let me log in. So, we're going to use Kimmy K2, which is a Chinese model. All right, we got search, we got thinking enabled, we got K2 flagship model, and then we're going to try and use this magic. All right, we have tested everything today, haven't we? We've tested uh Gemini, Claude, both alphas, Ernie, Stitch. So, let's see what Kimmy K2 can do as well. Mark says, "What's the best super agent model? " I would say Gen Spark is my favorite by far. I think that's the best model so far. Jason says, "Let's go, Kimmy. Fingers crossed. " All right, predictions before this finishes loading. Who thinks Kimmy is going to do the best? And who thinks that Claude is going to be the GOAT out of everything? Perang says, "Gemini 3. 0 RiftRunner is much better at coding. The controllers generated by is better than Claude. " Oh, that's impressive, man. And Paddyy says, "No, this is not Gemini 3. " They say if the if you use Gemini mobile app, canvas app, you get way better results in um canvas. Mark says Jensen Spark's the best and I think Kimmy would have centered. Claude for the win. I'm going to go with Claude. I think Claude is still going to perform the best. Let's see what we get. Peeps. See how we do. Mark says, "What's the best vibe coding ID app website? " My personal favorite is AI Studio at the minute. It seems to create like the best tools. It integrates APIs much easier and it seems to perform the best. Right. The sporter says Miniax. I don't want to sign up to Miniax on the subscription again, but last time I checked it was okay. Do you know what? I could probably try it in open router uh miniax. Let's have a look here. Is it on open router? Yeah, M2. All right, let's have a look. So, we got M2 over here. We'll do exactly the same prompt, right? So, create a PS5 controller. And then we'll go with interactive app. Okay. Create artifact. Boom. Let's go. All right. So, that's generating the background. And then if we go back to Kim K2, we'll preview this. What we got? Oh, it don't move. Yes. I mean, what is that about? Why is that moving? To be fair, I mean, it's done better than most of the other models. So, I'm not going to be too harsh on Kim K2, but you can't move the analog sticks. It still looks a bit weird. The UI is not that nice and the square button is moving around, and we don't know why, right? I mean, if you're playing FIFA, mate, that's going to be very messy. So, I still think that the Claude is a goat and it's beating all the other models. As you can see, you got the analog sticks moving there. You got L2, R2, um, everything is clickable. Pretty nice, right? Whereas Kim K2, it's done a bad, it's not done a bad job. I mean, it is a free model, so fair enough. But yeah, there we go. All right. This new model are garbage. I would agree with you, my friend. When do Gemini 3. 0 will release? When do you think Honestly, it's kind of crazy that GPT 5. 1 released first. And also, GT 5. 1 is total trash from what I've seen so far. Mini Max versus GenSpark could be a good video idea right there. Jason says, "There's something happening with the Gemini mobile app with a newer version than the desktop. " That's interesting. Peron says, "I would DM you a photo on X if I can of the controller by Gemini 3. " Okay, thank you. Yeah, feel free to DM me. Like, you can DM me on X. No problem. And then Mark says, "What's your go-to LM? " So, like I was saying before, well, I mean, to be honest, I'm going to go with Claude. Claude, I'm switching over Claude to everything now. Like I was using chat chy but I think 5. 1 like all right for personal task day today I was using chat chity 5 5. 1 came out it's a default model and it's really bad right it just it looks weird the response is weird it doesn't format stuff nicely so I think I'm going to try and train my brain and build up the habit of using claude every time instead and just try and get better responses inside claude and then for coding I like Gemini and Claude Minia did me good all right let's have a look what we got on Miniax so we've got the outputs over here. Let's see. But by the way, before we go on this, who thinks that Miniax is going to do a better job and who thinks that Claude's still the goat? I'd just be interested to know. Perang says, "I sent a DM on Twitter of Gemini 3. " Just going to go to Twitter. Oh yeah, that's pretty nice actually. It's not bad. I think Claude will win. I think Miniax will do good. I think Claude everyone a lot. Right. We got two votes for Claude. We got one vote for pair ranks. When do you use Gen Spark over Claude? GBT. most of the time honestly cuz Gen Spark's a little bit slow. So like you know the only thing that I'm going to use GenSpark for is like the bigger tasks that require deeper research or something where it would take me like 45 minutes to prepare. Droid with Claude best so far and then local go to local LM in LM Studio. I don't really use local that much. Um but when I did what was the one that I used by Google I got some training on it inside the AR money lab. It is what is it now? I forgot the name of it. it, mate.

name of it. Forgot the name of it, mate. Um, I think it was Gemma. Google Gemma. That was it. Yeah. Jason says Gemma. Yeah. Yeah. Alarm and LM Studio. Good with Gemma. Gemma is super fast as well. Let's see here. All right. So, we got two votes for Claude, one vote for Kimmy K2. Let's test it out. Go on the preview. There's nothing there, mate. There is nothing there. It doesn't make sense. So, again, Claude is a goat. K2 had a good crack at it. Miniax totally failed. And there you go, my friends. All right. Do you think Google's winning the AI rice? 100%. 110%. All right. Per this is I it doesn't feel like a productive use of a Sunday morning, but you know, I I'll do it for you. you guys. That That's a conspiracy theory, but I like it. Do you think Google is winning the ARS? Yeah, I do. I think like, you know, if you look at their products, they're just getting better all the time. They have a much better ecosystem and reach than Claude, right? So, for example, with Claude, yes, they've got a very good model, but they don't have the user base. Whereas Google, they already They already have people using Google and all these other, you know, they've got people on YouTube as well. Um, and so like they can easily market and get more people to use AI, right? So, like even if Claude's product is 10 times better, Google's distribution is 100 times better, right? And also with uh Google like their products actually really good like and they release a lot of free stuff. Perang says, "Did you go ever test out Gemini Go? " I didn't. And then Quen, I think we we're done with the Chinese models. Like we tried Miniacs. We tried Ernie. Yeah. And then Butchie Halfway beliefs in Alphabet. Yeah. I would agree with that. I think like uh Alphabet is just, you know, they've crushed it, haven't they? They've absolutely crushed it. So that's pretty much it for me, peeps. There was a Vey3 update as well, 3. 1 update. So, let's have a look. They're back with another 3. 1 update rolling out now on desktop. You can have multiple references. So, that was one thing I wanted to share with you. What else we got here? Flow with is something that I really want to try out, but I haven't got an invite code. Don't know if anyone's got an invite code, but I can get access. But, I've heard a lot of good things about flow with is like trending like crazy right now. It seems like some really cool stuff there, too. And then we've already covered this, but yeah, open router releases Sherlock and it isn't great. All right, so hopefully it's not Gemini 3. 0 because it's not that great. All right, so thanks very much for watching. If you want to get um you know, connect with me personally, feel free to join the AI profit boardroom link in the comments description. We have a flash sale on right now, so you can get 20% off forever. You get an amazing community. We have awesome members. You can see we run a bunch of contests, different challenges, etc. We have a very active community with loads of cool stuff going on and loads of people posting in there too. We have 1,700 serious AI members that you can join and connect with. Inside the classroom, you get all of my best training. So, for example, a six week AI automation masterass. You get my best playbooks on like email automation, Twitter automations, newsletter automations, shorts automations, AI avatars, etc. So, all my best trainings are inside the AI prof as well. You also get some training on like lead generation and social media, email and content automation, NA10 templates. I think we have over a thousand NA10 templates inside this section here. So, it's a huge amount of value there. And then also you can get access to the agency course, all our coaching call recordings and Q& A call recordings from previous our best and latest SAP updates and we update this regularly. Plus, we have the AICO automation and YouTube AI course as well. On top of that, you get coaching with us directly. So five times a week you get coaching. The content inside the classroom is exclusive and can't be found anywhere else. Mark says you offer consultation calls. So if you want it depends what you want, right? So you can book in a AI automation session link in the comments description and then basically this is kind of like a call before you become a client, right? So if you want us to do stuff for you, you know, feel free to book in a call here. If you need help with AI directly, you know, you've got the AI profit boarding where we have five coaching calls a week. So you can get consulting on those calls. Damian says, "Enjoy the rest of your day. Have a great day, too. " And uh Jason is talking about some interesting stuff on Sony. So, [clears throat] thanks very much for watching. Hope to see you inside the AR profit boardroom. If you want us to do AR automation for you uh from our agency, feel free to book in the call link in the comments description. And appreciate everyone who watched the call today. All right. Cheers. See you on the next one. Thank you. Bye-bye.