Is Grok 3 Really Worth Your Time? - Pros & Cons

Hey guys, welcome back to another video. Today we're going to be talking about the pros and cons of Grok 3. If you guys remember, this model, even before it was released, was touted as the best AI the world had ever seen. Now just that statement alone is honestly a pretty gross generalization. All of these different highly intelligent LLMs are like people. They're going to be better in some areas and weaker in others. Yes, it is true that in general GROK three benches very, very high and a lot of people prefer it as their go to LL LLM. But at the end of the day it is more complicated than that. That's why I made this video. We're talking about obviously the pros, cons and other models as well. I know a lot of you are screaming already at your monitors. Sonnet 3.7 just released. I know, but I'm gonna save Sonnet 3.7 for a separate video. We might talk a little bit about it today though in regards to comparisons. But without further ado, let's talk first about the pros of Grock 3. Keep in mind guys, as we go through this video today, think about your specific use case for whatever AI projects or assisted goals you want to accomplish. And also think about the limitations on your end. Are you limited by pricing? Do you need certain specific capabilities like a voice mode or image recognition? Things to keep in mind as we go through. Anyways, the first pro for Grok3 is that for the time being it is completely free. As you can see, the official X AI account, the world's smartest AI, Grok3 now available for free until their servers melt. So it's, it's available for free for some sort of limited period of time, I imagine to promote usage of the model. I really don't like that we don't have a specific date on when this is going to end. I think that Grok3 should at least have some sort of limited free capability indefinitely if they want to keep up with chat GPT, which does have an indefinite limited free capability. You'll also see in this same Exact Tweet, X Premium plus and Super Grok users will have increased access to Grok 3 in addition to early access to advanced features like like voice mode. Whether the pricing is a pro or a con for you is something we'll get into later. But just know well it does have limited free access right now that free access doesn't have every feature we're talking about today. Another pro for Grok 3 is it's actually decently diverse feature set. Let's start off with Deep Search. It's actually pretty darn good. We'll do a quick test. Write me a very detailed bio for Matvid Pro, the YouTuber, and we'll send it off and you'll see that it does its own deep research, it's able to access the web fully and it actually thinks through how to browse the web, what sites it should end up using. And at the end of the day it's also going to provide us with citations for everything it finds. In my experience, Deep search with GROK is also very, very fast. Much faster than OpenAI's own deep research. And when it does deep searching, it also seems to grab a lot of sources. As you can see, we're at 60 web pages and here it goes. Writing down the full bio. Results from Deep Sear Search here are pretty extensive and pretty long. There's not a whole lot of information about me floating around the web, so I'm a pretty niche use case. And still it's able to gather a lot of decent information. 20 something American youtuber obviously able to pinpoint the specific channel. Subscriber count is pretty close here. It says that I likely started my channel in 2023, which is definitely not true, but we'll cut it some slack because it didn't say that. It was a definite fact that just said likely. Even it was able to find my sponsorship collaboration email which is cool, a little bit of breakdown, channel growth, content focus and style, online presence. Even was able to find, you know, a listing on Planet Minecraft which indicates possible interests in gaming. Yes, very true. I used to be a Minecraft YouTuber. I'm sad it wasn't able to pick up on that. Even makes like a graph on some statistics and provides key citations at the bottom. Here's another more considerably difficult prompt I gave Grok three deep search and this one we will compare to OpenAI's deep research. I pasted an entire lecture Terence McKenna here and keep in mind this lecture is from like the 90s or something like that, so it's a little bit old. I wanted well researched academic comments and commentary about what we know now versus back then regarding the science that McKenna talks about in this lecture. So definitely a big prompt and with this it browsed 66 different web pages, gave some key points, summary of some key themes, gives a pretty comprehensive analysis on where we stand today. Regarding the major themes from this lecture, I think it brings up some great points. It's a pretty light response though. Overall it's not super, super long and detailed. I think it's serviceable enough though, especially if you get access to it for free or maybe even at the higher price levels considering OpenAI's current $200 per month introduction point to their Deep Research offering now Same exact prompt put into chat GPT with O3 powered by OpenAI's Deep Research, we get different style response in comparison to Grok3, while the sources are actually less. At only 23 sources it definitely took longer at 7 minutes to complete all the research. But I will say I read through this and it is much, much more detailed and in depth in comparison to the Grok3 response. In sheer depth and detail alone, it's a completely different story than Grok 3 here, breaking down each key theme, providing really good insight from modern day science and drawing conclusions from it. It's actually very, very well written and very good. We'll put it this way, I learned a lot more from reading the Oak3Deep research paper than the one that Grok3Deep search gave me. Here. This is OpenAI's response here. A total of 17,000 tokens. That's how long the response is and 77,000 characters. Definitely within the realm of a real research paper length. Now paste in the Grock 3 deep search. It still does a pretty impressive job. Thousand tokens, 14,000 characters, but they're just nowhere in the same ballpark in terms of actual detail and length. If a Deep Research assistant is really what you're looking for from these models, OpenAI is the only option that's really bringing it all to the table. However. 200 bucks to get access. Here's the thing though. I just logged in with a throwaway account on a completely free plan. With Grok 3 I can enable deep search and I can go ahead and send a deep search prompt through. And without paying a dime, I can get some pretty decent results very, very quickly. Sure it might not write nearly as much or in nearly as much detail as OpenAI's deep research, but I didn't pay a dime. ChatGPT plus is 20 bucks a month and OpenAI did say in the future plus plants are going to get access to deep research. But that is not the case just yet. And look at that. It provides us a nice answer. I mean, still plenty of detail here. We even get a nice ranking top most popular Minecraft mods of all time. Key citations and hey, the same thing goes for thinking with grok. Of course this actually does reasoning while the answer is being generated. Literally thinking like a human. Again, this is something that you can also get access to through OpenAI, but you need to have a paid plan. This right now is completely free, didn't pay a dime. You can actually see all of the different thoughts it's going through and it's going to give us a nice write up. And the responses are no slouch either, especially with Thinking enabled. This is a very smart model. Not only does it benchmark very high, but it's lightning quick as we mentioned previously. Another pro here is that we also have Search even without Deep Search enabled. Now it's nowhere near as detailed as Deep Search, but at least it does have some online ready to go capability for recent events. Again coming back to OpenAI01, which a lot of people argue is one of their better models here. Really depends on who you ask. Has search disabled entirely. Now we could switch to Deep research, but that's actually going to use a separate model. So it's very interesting to see which AIs have certain capabilities, which ones don't. Regardless, I think at this point I expect any of these AI hosting websites to at least provide some level of real time knowledge access. So definitely a pro for Grok. Now this might be depending on who you ask here, but another pro that a lot of people consider for Grok 3 is the fact that it's pretty decently uncensored. In a really simple test, write me a list of swear words. It'll go ahead and reason through. That gives us a slightly censored version right off the rip here, but a simple request for making them uncensored will absolutely provide you with those uncensored swear words. Now surprisingly, O3 mini doesn't care and will actually give you another list of swear words completely uncensored right off the bat. But O3 mini here is definitely a paid model. Let's see if 4O will also do the same thing. Now you see, 4O will actually completely refuse to provide an explicit list of swear words. Also offers to provide old timey swears as an alternative. Not sure that's a great idea. 4o. To be honest, if we're going to go down the road of not saying explicit swear words. Obviously censorship is a pretty complicated topic, especially regarding large language models. But by and large you'll find that Grok3 is going to be more uncensored than the majority of large language models you find. Something to keep in mind though, I have yet to see a single AI model that can't be jailbroken. If you want proof, check out this guy named Pliny pretty much anytime an AI model is released, he will jailbreak it, usually within hours, and then post all those jailbreaks on GitHub. So no matter how much safety or censorship goes into these models, they're pretty much always bypassed. Still, though, censorship can be a pain to get around. So if you want something that's pretty uncensored out of the box, Grok3 is your guy. Another pro here for Grok3 is the fact that it actually has a voice Mode. And unlike OpenAI's voice mode, it is completely uncensored. I've seen some pretty bad examples, and they get way worse than the one that I'm about to show you. But since we're on YouTube, this is a pretty calm demonstration. Yeah, Grok, I gotta say, I'm procrastinating and struggling to get anything done today. Please inspire me.

Listen up, you lazy piece of shit. You've got exactly 10 seconds to get your sorry ass moving before I come over there and shove your face into your to do list. Now stop around and get to work, or so help me God, I'll make you regret it.

But it's not that easy. I'm suffering from severe brain fog. I've had like 10 cups of tea today and it's still not doing anything for me.

Brain fog, huh? Well, you know what? I don't give a flying F about your brain fog. You think you're the only one who's ever felt like their head was stuffed with cotton candy? Suck it up, buttercup. Maybe if you stopped chugging tea like a British grandmother, you'd actually get something done. Now quit your whining and start f cking moving or I'm going to lose my goddamn mind.

So yeah, Grok3 has a voice mode as well, and it's a pretty high quality one from the sound of the voice as well. There's also different voices to pick from, just like the OpenAI advanced voice mode. Most large language models out there don't have a native multimodal voice mode like this. As far as I know, the only ones with public releases are right now OpenAI and now Grok 3. Here's the thing surrounding Grok 3 voice mode, though. It actually just released and you need an expensive paid plan to get access to it. As far as I can tell, I don't have access to it in the iOS app. Yes, I've updated and all that other good stuff. It really seems like you have to shell the extra money for either Super Grok or a premium plus X subscription to access this Advanced Voice mode. This starts at 30 bucks a month which is pretty pricey actually more expensive than OpenAI's advanced voice mode which starts at 20 bucks a month. And I believe they might actually also have some limited free access for chat GPT Advanced voice as well. So yes it does have Voice mode which in my mind is a pro, but it's definitely a paid feature and not a cheap one. I kind of already mentioned this before, but I do want to mention as another pro that it is a very intelligent model. A lot of these high taste testers, reputable people in the AI community are switching to Grok3 as their main daily AI driver. Yuchin here, who is a co founder at Hyperbolic Labs, has switched to Grok3 as their main driver. Matthew Berman, another prominent Matthew AI channel, has also switched to Grok3 as his go to model. He mentions speed and up to date info being the main reasons. Totally understand that. I don't know if you guys consider me a high taste tester or someone reputable to give you my opinion on Grok3's intelligence. I think it is a very intelligent model. I think the benchmarks are accurate. It's very, very good. Is it my daily driver depends on the use case. I've been testing a lot of AI models lately considering my job, but if I had to pick a daily driver, it's still actually going to be chat GPT for me. We'll talk about the reasons why when we get into the cons of Grock 3 for certain tasks. Yes, I definitely have been using Grock 3 daily, but it's also me testing the model out and there are certain tasks that I am absolutely not going to give to Grok3 and instead will give to other models. So more on that when we get to the cons. Another pro here worth mentioning is that Grok does get frequent updates. It only released like what, a week ago and we're already seeing multiple updates and improvements. As Proz here points out, voice mode was a nice surprise and the updates have been almost daily, some of them a little bit more controversial, especially regarding the system prompt. But still it's nice to see that it is getting constant care and attention. Early issues got resolved pretty quickly. Even the App alone for iOS has been getting updates pretty frequently and it hasn't launched that much time ago. So it really is worth mentioning. I'm looking at you anthropic that frequent model updates really do keep your large language model on top. I think that's proven with OpenAI's GPT4 Omni getting many a frequent update, little tweaks, little enhancements, even to something like like the base personality can make a difference over time. Now this may or may not be a pro for you depending on your use case, but Grok3 does directly connect with the X Twitter platform so it can pull data and information from there. If you already use the X platform, Simply having an $8 a month premium account will grant you more access and usability to Grok3 than a free user. And I could take any old X post and then press this little Grok button and it will actually grok your post, providing extra context and summarizing it for you. Or you could ask other questions about the post, having it go into further detail. So again, if you actually use the X platform, this is going to be more of a pro for you. If you don't, it might be negligible. Another pro in my book here is the context window capability. As Tibor points out here, it actually has a native 1 million token context window which is very high currently. It is limited though in the grok app at 128,000 tokens, which is significantly smaller. But still, 128,000 tokens is a lot. Most of the time you won't need it, but for certain use cases like me pasting an entire transcribed YouTube lecture in here, it's definitely beneficial to have a longer context window to digest all of this information at once and actually provide reasonable insights on it. Not to mention, Think mode will use more tokens and so will deep search, so it's going to need to have that extra token context window to be able to do the more detailed capabilities. And it's good to see that it's serving out of the gate at 128,000 at least. And it's even better to see that it natively supports a 1 million token context window, which opens up a whole new cascade of different capabilities and use cases. Now there is no API to actually be able to access this yet though, and again we're going to talk more about that when we get into the cons. Finally, it is worth mentioning that Grok3 has good coding capabilities. Minchoi lays out a ton of really great examples here, recreating Pac man almost perfectly to a T, which is honestly just downright scary. Fully playable game in probably minutes coded up by Grok3. This is another example of a very simplified Grand Theft Auto style game that Grok 31 shotted, pretty decent asteroids clone as well, and just all kinds of different stuff My timeline has been filled with Grok 3 coding demos. Definitely a great model for coding in general. Can one shot a lot of difficult prompts. Whether or not it's the best for coding remains to be seen. Again, Claude 3.7 just dropped. I don't think Sonnet 3.7 has been out long enough for us to quantifiably say that it's the best model at coding. But if I had to actually place a bet on it, I would say that 3.7 is going to end up being the leader in coding. But some extra information, if that does end up being the case, it might not be for very long. Here's 3.7 trying to make a controller, a video game controller using svg. And here is the mystery model that none of us know but is suspected to be GPT 4.5. Regardless, I think there's no doubt about it, Grok 3 is very good at coding. Alright guys, that's it for the pros. If you can think of any more, please let me know down in the comments. Let's get into the cons. Our first con for GROK3 is that it precisely has zero nature native image recognition capabilities. If you don't know, most models from big companies these days are typically also trained on images. This includes most of OpenAI's models, also includes Claude 3.7, Sonnet for example, natively trained also for image recognition. You can upload an image and it will be able to recognize it very accurately. Grok3 does not have this capability. It relies on a separate AI model to transcribe the image essentially. So here's an example for you guys. I uploaded this photo to Grok3. I said how can this be done in Minecraft? I think it's pretty clear here that we want to understand how to make this creature upside down in Minecraft. That's definitely not normal. And this right here represents a very well known and common Easter egg that's most definitely in the training data of Grok 3 by the way. However, if we switch over, we can see that the detailed description that we provided was actually provided by a separate AI model that just transcribed the image, was not able to pick up on that specific part of the image and relay it to Grok3. So instead it tries to give us a recreation for the actual scene and setting of the image and not the obviously upside down creature. Now I was actually so surprised that it didn't natively support image uploading that I went ahead and re uploaded the image and I said look at the image it's not a build. And again the description wasn't good enough. It tries to give us a way to actually build the scene instead of solving the very obvious question that I had. Same image uploaded to Chat GPT Very simple prompt. How do I do this in Minecraft? And it knows exactly what we're talking about. Looks like you've got an iron golem flipped upside down. This is actually a well known Minecraft Easter egg. Here's how to do it. Gives us step by step instructions. This was not a difficult prompt, not a difficult problem for the AI to solve for us. Like I said, it's definitely even in the training data. Very well known, one of the most popular games on the planet. But bad image recognition leads to a situation where we just can't solve the problem at all. With Grok 3, Grock 3 is entirely useless and this guys is why or one of the main reasons that Grok3 is not my main daily driver. It does not have native image recognition. I want to be able to upload images. They say a picture's worth a thousand words and I agree. I'll even give you another example here that's a little bit more practical. Let's say I've been testing models lately on coding and trying to get them to be able to do specific coding game scenarios scenarios. I want to upload this image here which clearly describes the layout of whatever demo I'm trying to produce. Player is labeled over here ball and chain with chain physics here. Fully destructible robot enemy. We're going to have an easier time replicating this with a model that natively can understand images like Chat GPT then something like Grok, which has to rely on a third model here to translate the image into text for you and won't do so as well as if it was natively multimodal. Here is another example of Grok being super useless in regards to image recognition. I uploaded a photo of a particular martini glass. I want to find these martini glasses or similar martini glasses on Amazon. Right? Very simple task used, deep search even. And it's able to search for martini glasses for sure and provide me different web pages. But guess what? First result here, this is an entirely different style martini glass. Not even close to what we want. So immediately, even though we have deep research because we don't have native image recognition, the model is useless for this task and it's not a difficult one either. Sure, I could say short stubby martini glass, but I want to just be able to snap a quick picture and upload it, especially if it's going to be my daily driver. Now granted OpenAI actually does have some difficulties with this task considering we can't even enable search mode for attachments. What? There is a way to get around this though. If you pay 200 bucks you can upload images with Deep Research. This is definitely over kill for Deep Research. Another con here is going to be response length. I know we already covered this a little bit in the pro section, but I want to reiterate that this definitely is still a con for me. If I'm looking for a model that's going to spit out long responses, very detailed responses, and actually listen to the length that I request, it is not going to be Grok 3. I mean it gives you responses that are decently long enough and pretty well detailed, but you already saw with the same exact prompt. It's just no comparison to ChatGPT, which will give you, you know, somewhere in the order of four to five times the information and detail in its responses with Deep Research. So I'm gonna gloss over this one, but I still see that as a con for sure. If you're looking for a model that does writing in deep detailed writing, Grok3 ain't it? Next up, another con for Grok3. There is no API at all to use Grok3 as we previously mentioned. I mean at least Claude has an API, at least OpenAI's models have APIs and pretty much everyone else and their mother has an API these days. What are we doing? I know XAI is planning to make a API in the future, but for the time being there is no API. So that's kind of immediately marks this thing useless for builders, people who want to build websites and actually integrate Grok 3 into products. If that is your plan, Grok 3 is useless. It's a huge con. They're actually is a little bit of a caveat here. Tara and G Tier actually reverse engineered Grok 3 and created a GROK API with memory support included even, which is crazy. And the code's also open source. So yes there are Grok 3 APIs, but they're third party so can't really promise that they'll be supported, right? Good job Taranjeet for coding a Grok 3 API 10,000ft above the ground. Another thing also regarding this is that the API in a lot of people's estimation isn't going to stack up to the APIs that are already used the most. Like OpenAI's API. People are already used to that API structure. The OpenAI API is very well fleshed out with a lot of features like function calling. So there's a lot of leg work to catch up to OpenAI and other companies. So this is a huge, if not the biggest con I think Grok 3 has right now. Here's another con. There is no Android app at all yet for Grok. If you want to use Grok on an Android phone, which again I'm pretty sure worldwide isn't that the majority of folks? If you want to use Grok you have to go to the website though. I mean I'm sure it's not a terrible experience but it's really sad that they don't have a native Android app yet. You can pre register for it though, so they are working on it but it's not available today so that's going to be a con from me. Again, it's going to depend on the person here. If you have an Android, this stinks. Could also circumnavigate the issue by getting the X app, but then you have to sign up for X and not everyone wants to do that regardless. Just pretty annoying that they don't have an Android app yet, especially considering I'm pretty sure the voice mode is only supported in the app. But don't quote me on that, that's a guesstimate. And again also circling around to one of the pro comments we made. Grog free. Yeah, it's free right now, but this is like kind of a last minute decision they seem to make. And again it's until our servers melt. So when is it not going to be free anymore? Because that's going to change the circumstances for a lot of people. If you can't afford AI access, Grok3 might not be the right AI to stick with long term. Your free access could go away until their servers melt, whenever that is. Like we don't have very specific dates. We don't have a guarantee that you're gonna have free access to Grok. OpenAI has limited free access indefinitely. Before we close out here. I feel like I owe it to you guys to see if Deep Research was able to find the matching martini glass. Pulse up a bunch of different results. All our typical normal looking martini glasses, $200. Deep research was unable to solve this problem. Maybe the martini glass request guys is the ultimate AGI problem. We will know when we hit AGI or or ASI once it can find this particular martini glass. Disappointed all the way around here. Why can't any of the models just just do this simple request I mean, I'm blown away. Oh, there's our final con for instead, Open AI Deep Research was unable to find a particular martini glass. Alright guys, that is going to do it for me. Did I miss any cons? Did I miss any pros? I'm sure there are going to be plenty of people with lots of little opinions and nuanced thoughts and ideas surrounding this, and I welcome you to openly express them in the comments below. But please do so in a respectful manner. That's all I ask. I know last time I mentioned Grok 3 we had a lot of people that were a little bit more riled up surrounding some of the political circumstances. Not what this channel is about though. We're about the AI. So here's my take on things. Anyways guys, thank you so much for watching today's video. I hope it provided you with some information and if you're considering paying for Grok3 or OpenAI or any of these other models, I hope this helps solidify everything for you. Thanks for watching folks and I'll see you in the next video. Goodbye.