I Tested Meta's Muse Spark 1.2 So You Don't Have to…

I tested META Muse Spark 1.2 so you don't have to and I'm going to show you how it compares versus GPT 5.6 Sol, we've got Fable 5 and 3.8 as well. This is one of the first builds that we've created with this. If you're not sure what's happened, basically META have just released Muse Spark 1.2 and Muse Code which is their agentic harness for Muse Spark and we've tested out already. Now on the benchmarks actually to be fair it was compared against for example Opus 5, GPT 5.6 Terra so like smaller models but I'll show you what we've built and what we got out of it so far.

So this is the first build here, honestly not that great compared to everything else so when I have a look at this so far I'm already thinking okay this is not going to be so good. If we have a look at this one this is super slow as well, super slow and laggy so it does struggle. It kind of feels like an older generation model when I'm testing out on pretty much all of these builds. This is like an RPG style game and you can see like even the camera angle doesn't make much sense however if we compare that versus Fable 5 it probably still did a better job there but I mean compare it to for example Quen 3.8, Quen 3.8 crushed this task versus Muse Spark and also to be fair GPT 5.6 Sol did a much better job.

So the thing to note here is like this doesn't mean it's a really bad model but it does mean like it's nowhere near the same level as a frontier model as well and also if you're going to build something out I probably wouldn't use it directly for building out stuff like this. There's some other stuff as well so for example like a flight simulator and it just feels like super buggy when you're using it compared to everything else. Let's have a look here we've got a Neon Blaster game. This one was not so bad to be fair but it just lacks detail when you're building with it.

Having said that it is pretty quick to reply as well. Let's take a look at the next one this is something we call Twilight Vale and when I try and move backwards it just spins around so like the controls are pretty bad there and again like games are just great for testing you know UI, gameplay, the detail, the vibe, how well it's reasoned and thought out the controls and everything else and this is just really really buggy like a lot of the builds that we're testing out here. This one was actually probably one of the best options that we got here so it's kind of like a racing game as you can see it just the only thing that I would say here is like it just kind of lacks the open world feel that you would hope from something like this. So if you look for example we can't really travel that far and it's also quite slow when we test out whereas for example if we have a look in comparison versus Quen 3.8, Quen 3.8 again like did a much better job and some people are even saying like Quen 3.8 is not that good but from everything I'm testing it is a great model especially compared to for example Muse Spark.

So look this was a Dragon Realm game it just bounces on the spot when it's moving around it doesn't make any sense. Let's take a look at the next one yeah it's just a common theme where it's going to be really laggy on most of the builds. This one was probably one of the better ones as well so it's kind of like a flight simulator game as well pretty easy to control it's got some cool stuff in there and the controls are nice actually did a much better job than Fable 5 first round as well but I would say GPT 5.6 Sol and Quen 3.8 did a better job at the same time as well and that's actually something that I'm seeing consistently is like it's just not winning on any of these tasks and the reason that we actually created Goldie Bench is just to test this stuff out yourself so you can see okay how does it perform how does it work um in reality you know not not against the benchmarks. Now if we have a look at this one look how basic the the game here looks it just feels really weird when you're using it it's not playable at all whereas like nowadays you can build something absolutely amazing that's open world and 3D like you can see here with Quen 3.8 Fable 5 nailed it GPT 5.6 nailed it as well now to be fair to it it did create something decent here so this is kind of like a Doom style game it looks pretty nice it's easy to navigate looks interesting etc.

This one was very very basic indeed it lacks you know it lacks a fun factor that's what I would say here and this is a crypt game but you can see when we move around like the arms are kind of missing let's test out running the character runs nicely and that's okay but for example like GPT 5.6 Sol did a much better job there let's see what we got here actually this one is not too bad it just like if we have a look at this wall or if we have a look at the details you see how that character has got stuck on the wall so it's not very good at like reasoning or thinking things out whereas if you have a look at Fable 5's version look at the lighting look at the colors look at the field vibe the shadows everything is is right there if you have a look GPT 5.6 Sol much more interesting and then Quen 3.8 created something really really nice there as well probably people watching this gonna think like I'm trying to push Quen 3.8 but I'm not it's just this is literally what you can see is it looks really good when you build stuff out with it the dragon flight game I don't even know how to change the camera angle here oh there we go super weird like it's just floating around not doing anything this one not particularly great this is super slow let's test it on the parachute game yeah it's okay but you can see it's not really moving down whereas you look at Fable 5's version how amazing does that look and then the graphics nothing else so overall is it going to replace any of the frontier models you use currently not at all MuseSpark probably not the best model I've never really been impressed with meta models honestly but what I will say here is that you can use the agentic harness so one thing that we've got for example is we've built in Muse code into our agentic harness here with the agent OS so we can use Muse code whenever we want and it can create like some interesting designs for websites and that sort of thing plus one of the benefits I think is that you can actually use it inside for example Hermes agent and if we compare this so we've got Muse over here and then we'll create a new tab and we'll test out versus Quen 3.8 and we'll just give it a basic task so we have Muse over here and then we'll select Quen 3.8 over here and we'll just give it like a learning task like learn from a guide for example so if we take this command here plug that in and we'll do the same inside Muse and we'll just compare okay what's the speed like how does it perform can it actually call tools like this agentically let's have a look and see how it performs here so you can see they're both thinking and what this task should do is basically learn from this guide so the forward learn command with Hermes analyzes a guide and then creates it as a skill so let's wait for them to load and it's actually replied faster than Quen 3.8 which is pretty impressive in itself let's have a look what the skill is like itself so if we open this up it's a pretty decent skill like quite quite in depth it's taking the information from the guide that we gave it analyzed it create a good table here being very specific with the skills and then you can see Quen 3.8 is still waiting to reply so actually you I think that you might not be using this for coding so much but actually using it for example inside Hermes or even delegating tasks to it as a sub-agent might be a better use of it because when you're using it directly inside the coder it's not getting anywhere in the same frontier levels however if you look at for example Quen 3.8 frontier model really really good for building stuff out not great for agentic tasks when it takes so long to reply here so that might be the best way to get the most out of it from everything that I've seen so thanks so much for watching if you want to get a full agent operating system we do have Muse code plugged in there so that you can use Muse spark you can code of it build stuff out preview it and then you can actually view what you've created earlier inside this system as well you can also use the Hermes agent os as well where we've got different profiles for different models we have the voice agent with Hermes, Hermes Oracle, Hermes Astros, a studio for building stuff out plus a full memory system so that the day that this new model came out we've plugged it into our agentic os you can grab that inside the AI Profit Boardroom link in the comments description or go to the AIProfitBoardroom.com inside the community you can ask questions get help and support in real time inside the classroom you can get all of our best trainings and courses and our full agentic operating system so we have a full zip file for it here and we regularly update it and then you can grab all of our new tutorials as well inside the calendar you can jump a weekly coach call share your screen ask questions and then inside the map you can meet people in your local area who are building out with AI agents like you so feel free to join link in the comments description or just go to the AIProfitBoardroom.com thanks for watching