All right. So in this video we will be comparing free AI models running on my laptop. Right? My 16 24 32GB laptop that you can run completely free private was this closed source private models like Tad GP Gemini and Claude and we are going to actually have them showdown and where each model stand in my tier list of S A B C D when it comes to these five parameters of content creation, emotional storytelling, user writing, and anti- AI slop and humanization which AI excels where and just to make this uh comparison fair. I will be using Gemini, Chad GPD and Claude and all of their free version and I will be using local AI models which I’ll be running I’ll be running using LM studio by the way you can do it check other videos on this channel right where I talked about how to do it but we’ll be using um models here like Quen like GMA by Google like Neotron and uh we will decide right we we we we both and we all will decide from my experience right and from understanding which one excels where and then what is the ultimate best way right so without further ado let’s get started and after the end of this video you’re going to have the answer of S A B C D tier which AI beats the current AI race all right cool so let’s start with the first test of creating a business owner focused content test right so I want to test if model can write a useful full content, right? Like LinkedIn post, social media post, YouTube post, and anything like that. And without sounding generic, right? So, I’m going to prompt it like this. Write a short LinkedIn post, blah blah blah. I’m going to run it through Gemini. It’s a free Gemini I just created. As you can see, it’s even giving me an option to upgrade. So, every tier, just to keep the comparison fair, I’m using all of their best uh free versions, right? So, I’m just going to paste it and I’m going to run it on Gemini. I’m going to do it. I’m going to do it on chat gpt and then I’m going to do it on cloud and I’m going to use sonnet 4.6 whatever it is. I’m just going to run it over here. Now I’m going to go over lm studio. These are the models that you can download that can run on your laptop without internet. And just want to give you a little bit specification about my laptop that I’m using right now to shoot this video and to run this models. So both of them um are running simultaneously. This is a MacBook Pro um um latest version with 32GB RAM. So ideally your uh your local model can fit under 16 GB 24 GB. The best is 24 GB. Ideal is 32. Best is 24. 16 is a little bit uh to fit. You go 48 or 64 GB RAM, right? Uh it works wonders. So let’s let me go inside the LM studio. I’m just going to command them, right? And then we’re going to compare and then we’ll put these things. All right. uh for which tier what and ultimately we will put the tier list. Right? So let me open LM Studio. All right. So I’m inside LM Studio and the way LM studio works is it’s a dashboard like this. I’m going to zoom in like this. I downloaded all these models here. I’m going to go to the chat function. Right? And I already just running some tests. Right? All you got to do in order to load any model I’m using the model, right? The model that I’m running, it’s a gamma 4 model. It’s a 26 billion parameter model mixture of experts 26B A4B quantised version so that it can run on our consumer GPUs. Now how do you download this model? You’ll just go on the left side here uh in the model sections here on the the fourth option and just at the top you will see or you can just show other stuff pics if you’re watching it later and then you will see this model. This is the model I’m running which is a gamma 4 new gamma 4 26 billion parameters mixture of experts with quantized version right with the training of quantization aware training so that it can run uh it can have the uh uh it can have the quality of like the actual full BS16 model right like so-called just so just so that to keep things simpler right so-called best models by using less memory that means while using less RAM so it can fit in into our consumer grade GPU uh GPUs and laptops right so what we’re going to do we’re going to go in the chat here and we’re going to load the model I’ve already loaded the model there is one issue uh that you need to do that there’s one not issue but there’s one thing that you need to fix before you run anything is you need to come here on the developer side and then go to info you will see the model that you’re running just make sure when you do the load and when you do the inference on the load side. This model is by default just to keep things context lesser. This this this thing should must be here around 4,000 or 6,000. This is a context length, right? That means how much in one chat you can fit in before it loses the context, right? So just max it out so that it has 226 to,000 tokens of context so that you can run it through multiple chats and it will not forget about what you’re asking. Right? So I’m just going to reload it to apply the changes. And then as it is as you can see it’s a 16 GB version. So so that’s why I’m saying 24 GB is a good uh RAM uh thing if you have but 32 is ideal and 6448 you just amazing right and running on Mac. Uh that’s why you get a benefit of unified memory though you can run it on Windows as well. So I’m going to run this here. I have loaded the model as you can see. Oops one second. I loaded the model and I’m just going to do I’m just going to run my prompt and then I’m just going to say enter. Now while we’re running this let’s see uh our uh you know uh our Gemini results discounts feed a transaction trust build relationship. When you’re a small business owner it’s incredibly tempting to slash prices to win a customer. We have all been there. But here’s the truth. Customer trust uh customer trust beats a discount every single time instead of lowering your prices. Focus on your raising reliability. Okay. Okay. Okay. Okay. Let’s let’s read the G GPT. Uh as a small business owner, it’s tempting to focus on discounts to win customers, but discounts get attention, trust, and loyalty. Customers come back when they know you’ll do what you promise. Uh okay, cool. Let’s cheat. Uh let’s read the um let’s read the uh the this uh Claude version. I learned this the hard way early on. Okay. I like this uh I like Claude gave me like two version story driven and a direct and punchy version. Let’s do the story driven. I learned this the hard way. So it’s more personalized. It automatically understood my uh like I am running it as a LinkedIn post. So it kind of gave it more personalized. I ran a discount, got a rush of new customers, felt great, then most of them never came back. I like this thing that what it did here if you understand you know what I mean right is like it kind of personalized it like as if somebody was running on scrolling on LinkedIn it would like have them like oh there’s someone sharing as opposed to the Gemini version or the charg version right uh it’s more of like you like like you are just talking about it discussing the topic in the while and here it’s actually driving through a story and the direct and punchy version also plot gave a discount gets someone in the door once trust gets them back. Oh, I don’t like that. I like the storydriven version. Let’s check the um uh you know uh uh let’s let’s check let’s check sorry the GMA version 4 this is by Gemini it’s actually free local we talked about this right so interestingly GMA also gave us two options which is I just which is what I like about it first of all just by seeing it it gave us two option the hard truth approach and the relatables. Oh, one second. It gave us option three. Okay, three options. Let’s let’s see if they’re actually relevant. Right. So, option one is the hard truth approach. Best if you want to sound like a mentor or a straight shooter. Stop racing to the bottom. Ah, interesting. It’s so tempting to throw a 25 uh one. Let me just increase it so you can properly see it. Option two. Let’s see the option two. The story version. I used to think that if I could not get a customer to buy, it was because my Wow. Holy moly. I spent way too much time worrying about discounts and competing on price. But I realized something. People don’t actually want the lowest price. They wanted the lowest risk. Whoa. Wow. That’s cool. They want to know that if something goes wrong, you’ll fix it. They want to know that you actually care. A deep discount is a band-aid. Trust is the foundation. Wow. Wow. It’s crazy. Uh, I love this version. So, I like Claude. I like GMA, but to be honest, I like GMA more cuz it’s a LinkedIn, right? It’s a LinkedIn post. I’m posting as myself. Why would I like this is a story version. I like it. And yeah, cool. Awesome. So, what are we going to do? We’re going to load another model. But so far, the models that we have tested, we’ll just rank them, right? So, so let’s do the Gemini first. So, we’re going to rank Gemini as a C tier. Chad GPD also will come in the C tier. I like Claude as a A tier and I would do GMA as A tier as well. And then let’s see how they rank uh across all the tests. Right. Two more models are remaining which are the Quen model. Right. One second. Yeah, the Quen model and the Neotron model. Uh let’s check them. Right. Uh these are also free models. I’m just going to copy my prompt and I’m going to go here and then I have all the models downloaded. So I’m just going to do eject. So it’s going to eject from my memory and then I’ve already downloaded this model. How to download the model again? Go to the fourth section and then just search for Neotron. And then in the Neotron I’m using the Neotron 3 Nano 4V model. Okay. You have nano3 omni omni model as well which is a 26 gigabyte model but I would do nano3 u sorry neotron 3 nano4v model. So I’m just going to use in the chat. I’m going to load the model. Uh the model is getting loaded. It’s a ggv version uh uh version of implementation where you know there are different optimization. We’ll talk about this uh as well right there are different way they optimize the version to fit in the consumer grade GPUs. So that’s the that’s the thing there. So I’m going to use this again before running it. I would just go to the developer version and then I will max the token length. If you see the max token length GMA was supporting um Gamma was supporting is 214 or something 268. This is only 104. That means you can’t run a long length query because if you run the long length query it will forget because it has a only 100,000 token context length. Right? But we’ll still run it. We’ll reload the model and we’ll see what happens. So, I’m just gonna That’s the only thing I I um I was thinking I’m going to run it on one of my one of my on Mac mini, but it won’t be fair so that we could do it simultaneously. So, just bear with me as we uh load and unload the models. Right. I’m going to give it this thing and I’m going to run it the same thing. Right now, while this is running, uh while this is running, uh uh the second test that is coming up, which is the content creator script writing test, it’s a script that I want to write for a Instagram reel uh with a strong hook and flow. So, this is the prompt that we’ll be using. But before that, I just want to read it. Uh I just want to read the Neotron version. Uh it by the way all the Neotron the Gamma and everything we’re running on LM Studio is running for free on my laptop without any internet on my device right completely private completely local why I don’t use discounts and why my customers thank me for it. Oo I I uh you know uh uh I like the approach it took. You know what I’m trying to say? like uh I I like what it took like why I don’t use discounts and my customers thank me for it. Nobody actually thought about this. Everybody was slamming discounts, right? Like oh you should not be using discount. Discounts will hurt you and all that. This took it in a very interesting. I like this. Last week I got a message from Sarah, my regular customer since 20. She said she brought our product because we didn’t have a sale. You always deliver what you say. She wrote that’s why I trust you. Discounts they’re easy but they don’t blind. Whoa. What’s your story? Uh oh, I like that uh um I like that uh call to action, right? Cuz uh most of these uh models they they they didn’t do the call to action. How do you focus? Drop your thoughts below. Actually, they did. But um uh they did, but uh what’s your story? I like just I like that approach, you know, like instead of asking, okay, what’s the one thing that your business does to build a customer trust? Like nobody wants to kind of share about that, right? as opposed to talk about me. Hey, have you been through the same thing? Uh what did you do? You know that kind of angle and approach, right? And then uh Claude did not do that. But still the Claude um uh had this attention grabbing approach. I learned it the hard way. All that, right? So Neotron surprisingly I like it. It’s a beam version, right? Let’s let’s check the Quen. Now Quen was actually So we just go to Quinn and then we’ll just go here and we’ll search for Quinn. You just have to download these model once, right? So, Quen 3.627B version. It’s a dense model that we running. I thought I was downloading it. Yeah, I think I downloaded it. It’s a big version. It’s a 92%. So, we can come and run these models uh run run with Quen later on, right? Cuz it’s at still 92%. I mean, it’s a 16 GB version. I don’t know how long will it take, but you get the point, right? We’ll we’ll continue with our test and we’ll run it with that, right? Uh so we’ll run all the prompt maybe later on with Quen just to kind of put towen as cuz we we have the context right we have we are human brains we have unlimited context right so that’s the thing I’m just going to run this now the the content creator test and I’m going to run with Gemini so I’m going to open a new chat just so that it does not mess up with the existing one and I’m going to paste the chat there right and then I’m going to go to chat GBT and I’m going to open up the new chat. New chat. Uh I’m going to clear the chat. I didn’t even log into chat GPD, I guess. Yeah, it’s just free, right? Just keep things fair and simple. Going to go to Claude and on the claude as well. I’m going to open a new chat. And then I’m going to run it here. And then also on uh here, I’m going to run here the new chat. And this is the Neotron version. And I’m just going to run this. And let’s see what GMA GMA gave me. Okay. Oh, Gma, not Gemini. Sorry. So, Gemini, right? You look directly into the camera, leaning into slight leaning in slightly. A bowl. Okay, that’s good. Here’s the uncomfortable truth about why you have not reached your goals. No, it’s not because your lack of discipline or motivation. When you start something new, a workout or routine, your brain gives you a massive hit. But by week three, the novelty veers off. in reality that boredom is just the tax you pay for the progress. It’s like your textbook. You know, you know what I’m trying to say? It’s it’s a it’s like textbook. It’s like okay, you know what I’m saying? It’s a I don’t like the hook honest. Right. Let’s see. Um I like Chad GPT uh so far like just on a second. I like Chad Gupt because Gemini I didn’t ask it to give me visual. Why do you why do you want to confuse me with the visuals? Right. I’ll ask you again. So that’s a Gemini issue. I always have that when I ask her for script but let’s read the script chat GPD. Right? Most people don’t fail their goals. They quit during the most predictable part of the process. I did not get that aha moment. Right? Anyways, let’s read through it. Think about it. At the beginning, progress feels exciting because everything is new. A few weeks later, the excitement disappears. You’re working just as hard, but the results are barely visible. And that’s exact moment most people assume something is wrong. So they’re still a textbook, right? I’m I’m trying to create a real that will go viral, right? I mean, I know I haven’t given them a context and there’s a whole context engineering role play where we give them a good context to understand from which I talked about in a previous videos, but I’m just because we are running a fair fair blank check. I no, I don’t like it. Okay, let’s let’s check this. Let’s check claw. Looks like it give it the hook. Okay. Uh, the day before something works, that’s exactly when most people stop. Ooh, there you go. Here’s what nobody tells you about pushing a goal. Progress is not linear. There is this period, could be weeks, could be months where you’re putting the work and getting nothing back. Relatable. No feedback, no results, no sign, and your brain behind a very logical organ. Think about like a pushing car. I like it’s a metaphor. It’s giving a metaphor. So, it feels relatable and throughout the field. I like this. You will let me know in the comments what do you think about this. But I like the Gemini and Char GBD is a more textbook textbook kind of approach. The CLD is much more like you know what I’m saying like u uh it’s more it’s much more uh uh you know what I’m trying to say like uh relatable and hookwise. Uh I I like that. Right. Anyways uh Neutron let’s see. Oh, by the way, uh some of you might be wondering, oh yes, I could hack away in Gemini or CHB. By the way, this is not a video where I’m saying this model is great or that model is great and I hate this model. No, no, no. It’s a fair blank [ __ ] comparison. So, you could say that okay, by doing some using some custom GPT or Gemini gems, you could improve something. Yes, you can. But ideally on a standalone model itself, which one is great? Right? So that’s what I’m trying to uh get here at. So another thing that I want to do here is by the way, oh Quinn is also downloaded. So I’m just going to go there. All right. So um I’ve been just like, you know, just go on on and on is loving it. Anyways, remember that exact moment your goal just stopped feeling real. Mhm. Okay. Okay. Like it was heavy, not exciting. That’s why 80 that’s why um that’s why 80% of us ditch them too soon. Not because we are lazy, but because we forget why we started. Oh, goals are not about checking boxes. They’re about person you become while chasing them. When that connection fades, you quit before you have seen it. Because goals without heart just burn out. Your why is not one time spark. It’s a daily heartbeat. Okay, I like it. I like it. I like it. I like it. I I would not give it a like a better than Claude, but Claude first, then emotron, then other models. Let’s check the other models. Like let’s check our J Gamer model. So I’m just going to load gamma. Let’s use the let’s load the gamma model while it is running. Let me make sure that the gamma model. So as you see what I did, right? I went to the developer documentation and I just loaded. So it’s it has a 262k tokens. I’ll load this. I’m going to reload to apply changes and then let’s see. Uh bear with me guys, you know, uh because uh this just one time thing. I want to do it on a fair device, one device. So then we know like you know I’m not running one device on 120 GB RAM and one on a different RAM. Anyways, so I’m going to do this and I’m going to start here and let’s go GMA. Let’s go. Awesome. In the meanwhile let’s just rank from so far what we have written. Now I would say from what I have written Gemini right it’s a very textbook thing. Uh it’s a very textbook thing. I will I will arrange the Gemini again for the third test will tell us better. Chat Gibb is also textbook. Claude is better. So I would I so two test it pass. I will I I will I will put Claude as a S tier. Let’s see what um uh what GMA is doing. And we’ll do the Quen one as well. But Neotron let’s see Neimotron version that we just saw, right? The Neotron version that was also uh that was also nice. But anyways, let’s see this one. So, GMA coming from Google did the same thing, right? Which is most people don’t fail because they lack talent. They fail because they quit right before the math actually starts working out. When you start a new goal, whether it’s a fitness, a side of there’s the initial rush of dopamine, but then you hit the plateau, you’re putting in the work, the scale is not moving, the views aren’t increasing, the bank looks the same. Real progress, it’s okay. uh you know uh you know what I’m trying to say. It’s okay. I would not say it’s too great. I like the Neotron version honestly, but let’s see. Let’s try the Quen version for this one. Right. Let’s try with the Quen version. Quen also has a 262,000 of token length. So, I’m just going to reload Quen. Right. Let’s see. It’s a 3.6 Quen 3.627B 6 27B model. I could run all these models. I could do all this. I want to do it in front of you. So, you know how I also, you know, think and then how we kind of decide so that not bias towards anything, right? So, I’m going to do Quen. There’s still one Quen uh that needs to be run. But uh anyways so yeah while we are coming here Gimma just because it gave me the okayish response on the real script for me it fell at B honestly Gemini and Char JBD has a very strong chance to go in the D level but let’s see A level I would not put Neotron A level but Neotron has a very strong signal of understanding the context and giving it it’s a small model this model the Neotron version that we we used actually this 8 A4B version it could even run in your 16 GB RAM. So considering that small model and understanding that uh that grade uh you know it’s a good good I a good thing right so that’s the thing and uh make sure if you’re loving it so far uh drop a like subscribe and then uh we will be doing basically you know a more version of building games application web apps videos and everything so we can see like which which one stand but claude for me it 100% stands is s tier both of them work perfectly uh gamma one uh the quen one is definitely taking its time cuz it’s a 27 billion parameter model that we using and it’s a dense model that means for example just while it is working just uh just so you understand what’s a dense and a mixture of experts model means dense model means it’s going to use all your RAM right mixture of experts model just for simpler simpler uh simp simplicity it will only pick the specific experts so your RAM will not get utilized all of it right uh so yeah that’s a simpler version, right? So, okay, most people don’t quit because they’re lazy. They quit because ah, come on. It’s okay. You know, they quit their measuring progress wrong. We pick a goal, expect a straight line. Your brain is not wired to reward invisible progress. I like that state sentence the inside. So, stop tracking the outcome. Track the signal. No, I didn’t like it. But still, I’ll just put it at D. I don’t like it. It’s a plain textbook version. So far, you’re going good. Let’s do the third test. Now we’ll be doing fast fast test. All right. So it’s a emotional storytelling test, right? I’m going to do Quen uh not Quinn Gemini. I run the test. I’m going to do Chad GPT new chat. Run this test. Do chat uh the Claude Claude new chat. By the way, I’m running Sonnet 4.6. And then uh I’m just going to run uh Quen here with the emotional story test. And then we’ll run Neotron as well. Um so let’s read through that. All right. So this is a good story overall. Uh again and u you know a lot of people might feel this other way and that’s what I would love to have your feedback or your opinion on what I’m thinking could be wrong right so you can share your like okay yes I could have thought that way as well. So anyways let’s read the emotional story emotional test. For seven years, David’s world was measured in lines of code and the hums of the office server room at midnight. He was back in software engineer for a massive healthcare app. The guy who kept database secure the data flow and again the textbook, you know, it’s like a textbook. It’s like I would not want to read this, right? I just read read a like paragraph. It was thankless work. When the app runs slow, nobody noticed David. When it glitched, his inbox blew up. While the marketing I related it like you know and I’m relating so far I mean till now this while the marketing team celebrated millions of user milestone with champagne one runny Tuesday a companywide email it was not praised there it was an invitation of volunteer impact panel too texty too I don’t want to read it something that I don’t want to read I would rate it like too texty too textbook for 12 years this is by the way Chad GBD for 12 years Ravi unlocked the office doors before sunrise he worked as a maintenance technician in a small manufacturing company. Every morning uh every morning checked the machine, fixed loose wires. When production ran smoothly, nobody noticed him. When something broke, people only asked why it was not fixed. Over the years, we watched younger employee get promotion. Okay. Okay. Okay. Okay. Okay. Yeah, you get the point. I’ll just in case if you want to read through that, it’s just not textbook, right? Like I might be biased here now by just reading like two paragraphs. Not biased, but I might be judging them early. Uh but it’s very textbook, right? I does not feel like reading it, right? Let’s shake the rod. Okay, the last row. Margaret had been the last had been the last one to leave building for 11 years. Every evening she gathered the forgotten coffee forgotten uh coffee cups straight in the chairs, water the phone with the window. Nobody me nobody mentioned it, not once. Her name was rarely in the meeting. Her suggestions arrived a beat too late. Um uh there were evenings she sat in her car parking not ready to go home not sure what she was waiting for does does it matter she would think does any of this actually matter I’ve been meaning I have been meaning to write this for a long time Priya wrote I’m almost quit Margaret read the email three times and she pressed her fingers by the way I’m just speaking but just so you guys know right you know I’m also speaking I’m also reading it through my eyes as well so whatever I’m speaking is not what I’m reading I’m just reading grew to understand, right? Felt like relief. She wrote back four words. She wrote back four words because it turned out it had mattered along. Okay, cool. Let’s check the Quinn. Uh, Mara’s knees clicked like a metronome 5:00 a.m. Rain or shine 480. She walked the three blocks. The neglected lord behind the bakery. There was no budget, no committee, just mara. She was put into the world. She was just the garden lady. a blur h like there was a no budget I just I’m just first two paragraph like what am I reading she was just a garden lady you are here late again you would grumble there were nights then a small hand tugged her sleeve they were not passing by she was not she was the root I mean okay I didn’t like that much let’s let’s try other model eject this and then I’m going to run gamma let’s try gamma so Far uh not I mean anything have to do writing GMA and Quen not good. Uh GMA and uh not GMA sorry Gemini and Quen I would not rate them. Chad GPD okay it was okay right. Um um uh like story at least I would keep it as Ctier. Let’s now test the final the gamer version the Neotronone version so that we can have an idea. Let me check if the context are good. Let me just raise the context. So this is a context length so that you know it can give us the full even if it continues to write continues to read it has it can you know uh it has much much higher context length. So I’m just going to read that go here and I’m going to paste that thing. Let’s see. It’s much faster by the way you know these um um contise versions uh contis versions mixture of expert models which is 26 billion parameter active 4B. Does that make sense? That means at any given time there are only 4 billion parameters or whatever it is it’s active. You see what I’m trying to say? That’s the thing with those 27B model um uh uh um you know it was a dense model right? All of the parameters were active. That’s why it was slow. So the quen one we just um um you know uh that’s the thing. So it it gave us a note. So for 12 years Arthur’s world was defined by the things people looked past. He was the man in the faded blue jump suite. The man who arrived before the sun to scrub the floors of the municipal library. He knew the scent of old. He worked in the silence of the stacks repairing broken spines. Uh dusting the forgotten uh some mornings when the damp winter chills went oh interesting. was a heavy gray. See this it’s very very very interesting Gemini right this is Gemini for seven individual line of code see there is this feeling right where Gemini and Gamma from the same company Google and Alphabet right but I don’t know but I’m liking what you guys thinking right I’m liking the GMA responses much better which is free by the way much better than the Gemini version Gemini is also nice but It feels like you know like the word look at it’s it has a story right it’s talking about pronunciations are there there’s a story there right but here right it just felt like you know like a textbook you know what I’m trying to say right it’s a very not so like how humans would write right like you know what I’m saying but here he worked in the silence some mornings when the damp winter chill uh then came a Tuesday it was heavy gray afternoon you know like like heavy gray afternoon just the depth moving inside the depth of that right Arthur was finishing his sip shift wiping down the heavy oak cables. Arthur paused his clothes held midair. See that those that that uh deeping down and then exaggerating and then you know deeping down their pain points and then not the pain points but actually getting into the root of the um um u uh of the story as opposed to like 3 years ago um you know we live in a rural town. She helped her phone. Memory flooded back. It’s very textbook, right? So that’s nice. That’s nice. GMA. I I like GMA. So the though GMA did not pass our real test, right? But uh GMA is nice here for the storytelling. Let’s look at the Neotron. Surprisingly, Neotron, I am loving Neotron, right? It passed the though it it has only 104K context. That means if you try to chat back, like if you ask it to write an essay and if you try to chat back, you know, it will have an issue because it can only fit in 104,000 tokens, right? Which is huge, right? Uh uh but still just want to let you know. And also the Neotron Nano 4V version is not a vision model. I’ll tell you what vision model means. Um I’ll while just writing it. Vision model means uh you can give it an image and it will give you the prompt back and it can understand image as well. This does not right. So um so let’s do the uh Neotron right? So Neotron it’s a small model right? Okay cool. Lita had scrubbed the same uh same stainless uh steel sink for 23 winters watching steam rise plates clatter away into oblivion. The dinner’s manager, the the diner’s manager called her the machine, never to say thank you. Her husband’s voice aloh in the background was only appla was only applause she ever heard. She fed her son Maya. She fed her son Maya on leftovers. Dreaming that one day his Okay. Okay, cool. uh dreaming that one day his success would um echo the uh gratitude she long denied. When when Maya sent home a college acceptance letter, Lena uh felt a weight lift, not from belief, but from a strange quite pride. Mom, I’m going to be engineer, she whispered. That night, a postcard arrived. No return, just her name. Your hands cleaned our hearts. Thank you, Le. Ah, it’s it, you know, I am not satisfied honestly. Right. The only ones that I liked here for this test is the gamma version and the claw version. So I would say gamma. Right. I would still keep Neimotron here only but I would just put it beside because Neotron succeeded at my real test but Gamma did not. But Gamma and Claude has the understanding of emotions and all that. Gemini playing textbook. Quen plain textbook. Chad Chipy okayish. GMA has a strong contender. Neimotron is also let’s try two more tests remaining right uh fifth one is the anti-AI swap and the fourth one is the user writing birthday message and stuff like that let’s do the anti anti-AI swap first right I’m going to do that open up Gemini right this oh by the way just in case uh you might be wondering why are you not running GLM from ZAI or DeepSeek or or Miniax Mistral and all these these models because I’m using only local models that could run in your computer in a local computer or local device those Kimmy models will do the test right those Kimmy models those GLMs those other things can run does run but the problem is um uh it does you require 100 GB of RAM or 128 GB of RAM minimum 80 GB 90 GB that’s why they’re big huge models right though they are open source we’ll talk about them later as So, so that’s a fair comparison like what every one of you can run. Uh, that’s the only thing we’ll use them for the white coding test, right? So, I’m just going to run it here. This is Gemini. This is Chad GPT. This is Oops. This is Gemini. This is Chad GPT. And this is CLOT. All right. While this is running, I’m just going to do the version here with let’s do the Nvidia one. anti anti-slop neotron and then we’ll run we’ll run the the we’ll run the the the quen right okay write a short uh so this is the anti- AI slop test right and there’s the everyday user writing test that’s remaining so write a short paragraph avoid learning from failure make it sound like a real person wrote it avoid phrases like journey unlock your potential okay honestly messing up sucks and it usually feels pretty embarrassing in the moment but if you can’t get past the initial sting failure is see again Uh, again it’s a textbook, right? Just aggressive way of showing you exactly what doesn’t work. You can read all the advice columns you want, but nothing actually sticks. It’s like, you know, it’s like I still remember my ninth grade teacher teaching me an English lesson. You know, it’s like failure can be frustrating, but it often teaches lessons that success does not. When something goes wrong, it give us a chance to see what needs. This is the anti-AI slop. So, understand that. That’s the that’s a that’s a um that’s a theory that’s that’s a perspective you’re keeping in my mind, right? from the from then the achievements instead of seeing failure. Okay, failure has a way of teaching you think that success never could. You think that when something goes wrong, a project falls apart, a relationship end. So you see this is this is what I meant by that. Right? So honestly messing up sucks and it’s usually pretty uh this is the Gemini, right? It feels pretty embarrassing and but if you can get past the initial sting, failure is basically just aggressive way of showing. You see pretty typical textbook. See the plot it talks about it. It it goes deeper or second level deeper, right? your relationship ends badly. A bet you were so sure about does not pay off. You were forced to sit with the uh you know in the job that you don’t like. That discomfort where the real learning happens, right? That’s what I’m looking for, right? Although honestly um I would give just one praise that the Gemini does not did not does not give you like the the M dashes though claude gives a lot of M dashes which is a one thing that is a it’s written by AI whenever there’s a m dashes so but that’s okay I mean it’s okay just thing but I like uh claude charge is also nice but again then something goes wrong charges is like you know not average daily anything like that so Okay. Ooh, Neotron always uh uh shines up. Look at this. I’ve learned that the biggest mistake don’t just feel painful. They also point out what I did not get right. And over time, those little clues add up real progress. When something goes wrong, it’s not dead end. It’s just another step. It’s like it’s a motivational speech. It’s okay. It’s not that aha moment what I got from the real, right? Um uh and also for two tests it did not pass the emotional test and this test. But anyways, let’s try the Gamma from Gemini. I mean not from I keep saying Gemini, it’s from Google. I’m going to reload the models. Let’s load them. And then uh I’m just going to keep keep them loaded. And then uh let’s see. Let’s see. Keep them loaded and again as you can see here vision right it’s a vision model you can see it’s a multimodel we call it right let’s see all right so far so we tested so far what’s the list is looking like right so we tested Gemini Quen okay chity is a Ctier okay tier you write it uh Neotron is coming kind of charad if this next test we do and it fails Neotron will come here the only test that I like Neotron is the the actual content like a real writing test but it failed emotional writing test. So basically it failed an emotional writing and the anti- AAI slop. I mean it’s still like it was not it was not looking like I would say if I was just comparing based on AI swap I think it did not it was not AI swap right it’s compared to what Gemini was like you know um but uh still what I was expecting was not there but anyway still considering it’s just a 4 billion model I would keep it I mean if this fails the next test I mean if if it does not satisfy me I would put it as a Ctier but I think Gamma and Neotron is the best competitor let’s see who ends up E, right? So, let’s see. All right, cool. Uh, I used to hate messing up it. See, this is what is crazy. Check this out. This is Gemini. It’s from Alphabet. It’s a Google model. Same local model is a Google model. But somehow it feels so good on local. So interesting, right? AI is sometimes shocks everyone. Okay, so it’s a game Google, right? I used to hate messing up. It felt embarrassing and honestly just felt like a waste of time. But looking back, the moments that actually stuck with me were not the wins where everything went. Exactly. It was the times that completely flopped. When you fail, you’re forced to stop. Man, this has to be a tyier model. This is a eight-year model. like like same company a paid model is actually beaten by heads and toes um by a free local model. AI always AI always um you know uh surprises us. Let’s test the Quen model. Uh I don’t have high expectations. Don’t expect something amazingly great. By the way, if you’re loving it so far, like and subscribe because what we’ll be doing, we’ll be doing prompt writing tests, story writing, storytelling, not just storytelling more uh you know, white coding, uh like you know, building applications, oneshot test and stuff like that. So, you’re going to love this. So, I’m just going to run the quen and there’s only last test remaining. While we are talking, let’s just let me just run the last test. The last test is supposed to be a everyday writing. So how how it can write the everyday you know like you want to write a quick message to your employee to your colleague to your friend how how does it do it right so let’s do Gemini man I I lost my hope on Gemini like well let’s see let’s see so I’m just going to do Gemini chat GPT uh chat GPT uh stay logged out I don’t want to log in I want to keep using free version and then uh I’m going to do this And for by the way for Gemini, I’m using the flash just so you know. It’s supposed to be the best 3.5 flash all around. Right. Anyways, let’s just look at our antislab test for Quen. Let me just zoom in a little bit. Zoomed in a lot. All right, let’s see. But in the meantime, let’s do the postal image. The final test. Happy birthday to my big brother. anyone could ask for. Growing up and even now, you’ve always been the rock of this family. Whenever things got tough, if someone needed help, you were always there. Put everyone else before yourself. Thank you for always being out say our safety net and for leading the way with so much grace. I hope today bring I’m sending my friend, my brother, a happy birthday message. But why does it sound like a textbook? I mean I am to be honest I’m a little bit biased with Gemini but still it does not feel like the emotion that I’m trying to you know happy birthday brick brother you know even if if what I’ve just given that I mean I’m like you know he he he will tell me that hey bro like you know you you wrote it on charity you could not even thank me with your own words I mean anyways I like it’s okay it’s not not that great ba oh this is oh that’s one interesting thing I liked about chachi somehow chhat chipity is giving getting access of my location a thousand%. It is always personalizing with Indian names, Indian slangs. So just be aware Chaji might be accessing your location data cuz how come without me logging out it can understand obviously these people through my IP address they can understand but but it is personalizing me. So if you would like you could if if you’re using it from like Europe, right? It would use the European slang and all that. But yeah. Okay. So let’s see. I just want to thank you for always being there for all of us through every challenge and every happy moment. You’ve been a constant source. I’ve always admired the way. So yeah, it’s okay. It’s nice. It’s it’s better than Gemini. Honestly, better than Gemini. Who talks about uh like this is a very textbook like it’s like my ninth grade teacher is wishing me happy birthday. You know, she also does it better. I don’t remember it. Anyways, um Claude, happy birthday. Bye. Okay. Happy birthday by I don’t I don’t say this enough but watching you carry relatable I don’t say this enough but watching you carry so much for this family quietly without I love you I just feel someone is telling me happy birthday has shaved who I am but I wish Claude if you’re listening to this like come on this M dashes I don’t like them they look weird and everyone knows that if there’s an M dash is a AI right we have been the steady one the one who showed up when it mattered who made sacrifices they were never asked of you but you were always noticed today is yours not the families not anyone else yours hope you feel as good as lovely lovely lovely lovely let’s just do let’s just do the last test with the anti-slop and then we’ll come to this happy birthday uh thing with the local model so let’s do the quen for the anti-sloping honestly messing up uh used to make me want to hide under a blanket there’s no getting around how much it stings when you blow something but I’ve learned the Uh fail is just expensive data. If you don’t trip, you never figure out where the rocks are. When I screw up, I try to stop pouting and actually look at what went wrong. Okay. Not go like not not okay. You know, I would not say like it’s okay. Quinn Quen for all the writing test.
Yeah. DD. No, it’s okay. All of the test failed. It’s a 27 billion parameter model. No way. Right. So let’s do the local last local model test last which is the user writing happy birthday. Let’s do the quen. Let’s write it. Uh you know um let’s see let’s see uh guys it’s crazy right? It’s crazy when you go down the rabbit hole the model that’s everyone uh is is claiming or supposed to praise does not perform and the the the the the models that are hidden behind the plain sight. the moral is supposed to not work or whatever you want to say right um it’s doing relatively well and this is this is the why this is why I don’t like to 27B for this writing purposes it overthinks it’s so it think like why do you want to think so much right uh first of all it’s a dense model but it over it overkills at thinking right I don’t want you like I know I could like you know uh turn off the thinking thing but it just thinks so much and gives gives me the gibberish answer. Um you know so uh we we we’ll test these things on ideiations as well, business ideas, video scripts and all that but just the overall test right which is the content creation posting storytelling normal writing anti-slop. So far it’s looking great. Let’s see. Let’s see who takes down Claude. Let’s see. All right. So happy birthday brother’s name. Wow, I’ve been thinking about you and the way you always looked out for all of us. You’ve been a you’ve you have you have a habit of stepping up when things are hard and quietly. I like this. You give me like three options. I didn’t ask for it. Happy birthday. I don’t say it enough, but I really appreciate you. Oh, that’s nice. You’ve always been the one we could rely. I don’t say it enough. See, that’s the emotion, right? Happy brother, the best big brother anyone could ask for. It’s nice, but you know, it’s emotion when you say like I don’t say it enough. This like feel of emotion. So it’s okay good model but let’s see we’re it’s okay I would not say like wow it’s okay good right uh we’re also comparing with the speed right speed of everything as well so I’m just going to make this model Neotron uh nano it’s a Neotron nano model and then it’s a 4B model the only reason right uh the only reason I’m keeping it at a B level is first of all it’s not a multimodel that means tomorrow if you want to put some images, get some prompts out of it. You can’t do that. That’s only the reason I put I’m putting so far at the B. Another thing, it has a lesser context length, right? Which is only 104K context. That means if you just go for continue to a long horizon task on back and forth, it will get out of the context and it will forget about it. So that’s why I’m not going to put it. That’s the only reason a quen is the quen is not that and the D level just for this quen is at the D level. First of all, it overthinks. Second of all, I don’t like that writing approach. But hey, happy birthday, brother. From the moment you’ve stood out, uh from the moment you stood out through the every high and low, I’ve seen how your study uh presents turns ordinary. Oh, typical happy birthday message. Um okay, Neotron, good job. I’m okay. I’m not surprised. Not surprised, but uh I’m okay with it. Like I’m okay. Not no complaints. No complaints. No complaints. Much more complaint. Um, all right. So, I’m just going to load the I’m just going to load the the the the GMA model. Let’s see. Let’s see. Let’s see. It’s It’s crazy. Let’s see how much GMA GMA. I have high expectations. Oh, my battery is about to die. Wow. It’s crazy. I don’t know how long I’ve been um running. Let’s see how much battery I have. Oh, 10%. I got to hurry up. Hurry up. It’s been almost 15 minutes I’ve been uh doing. But we are about to wrap it up. We are about to wrap it up, right? Uh see Gamma is also thinking but it’s thinking very fast. That’s what I like. Uh I like it. So say op GMA. GMA gave me four options. Happy birthday. I don’t say nearly enough but I see everything you do for this family. You’ve always been one we can count on. But I want you to know how much that means to me. Oo. I hope you have a great day. You’ve definitely earned it. Happy birthday to the guy who always has our backs. Thanks for everything you do to keep this family moving forward. We are so lucky to Even this is good enough, right? like you don’t I don’t want like something coming from your heart obviously like birthday message should be just a voice note happy birthday really appreciate you thank you so much for doing instead of this but okay happy birthday looking back I realize how much you have stepped up for all of us over the years happy birthday thanks for being one we can always turn to things when things get tough lovely you do so much for this family and we all appreciate you more than we say lovely lovely lovely so here’s what I would say GMA GMA man GMA salute GMA like it’s absolutely like I didn’t even ask for four options it’s ran very fast so here’s where we stand claude it’s a paid model though I use it for free it will obviously limit me right continue to continue with me this is the series where I’m comparing each and every model right so this is the writing test emotional writing and all that here’s the stat claude and GMA cla is the winner when it comes to if you want to pay for cloud and then subscription and everything when data steals on their server. GMA exact model GMA 426B A4B font uh version best model the only test where I didn’t like it um right is with the the emotional part of it right um uh uh but yeah I mean still right uh everything it beat it’s better than uh Neotron chat GPD is like okay if you want to get it something quick right Um um and then uh Quen I don’t like it. Gemini Gemini I I kindly request you not use Gemini. It’s a plain textbook uh thing, right? So Quen not happy. Chad GPD okay if you want to use something quick like something like you have memory there. There are some other ad benefits you get with Chad GPD. We’re going to talk about that. But here’s our winner. Claude GMA B2B a version. Nobody came with the A version right? Neimotron is a nice small little model works best right but again does not have image capabilities as well so we’ll try other Neo Neotron models as well down the line when we testing for that use cases chat GPD is a okay model you’re you’re on train you’re on gym and you want something quick it’s okay uh Quen um thinks a lot needs 27B it’s a 77B model requires minimum of 20 24 GB RAM not that great Gemini stay Anyway, I mean stay away for writing purposes, right? So, I hope you love this showdown. Got to hurry up. I charge my device. But if you do, please let me know what you think about it. What could be improved from this uh showdown? And I hope I’m able to uh come and share some interesting idea, perspective, and thought process of mine to for your workfl in in your AI journey. If I do, if I did, let me uh let me know in the uh let me know by liking this video and let me know in the comment sections. I’ll see you in the next one with this another amazing showdown for different purpose. All right, see you. Love you. Bye.