The video compares Claude Sonnet 5 and GLM 5.2 side by side using one-shot builds (dungeon crawler, raycaster maze, multiple games, a website UI, and a Web OS), noting Sonnet 5 sometimes looks smoother but often feels basic or lacks gameplay, while GLM 5.2 is frequently more interesting and polished, though it can be buggy in some tests where Sonnet 5 wins. It also reviews benchmarks (CursorBench and GaudiBench/Goldy Bench), stating Sonnet 5 scores higher than GLM 5.2 on CursorBench but Opus 4.8 outperforms Sonnet 5, and βFable 5β leads overall and is expected to return within 24 hours. The creator highlights pricing differences (Sonnet 5 far more expensive than GLM 5.2), GLM 5.2βs open-source and OAuth/agentic compatibility, demonstrates plugging GLM into Claude Code via Agent OS, and promotes the Agent OS and AI Profit Boarding community with tutorials, coaching calls, and support.
00:00 One Shot Showdown
00:18 Dungeon Crawler Test
01:07 Raycaster Maze Faceoff
01:52 CursorBench Rankings
02:28 Opus vs Sonnet vs Fable
03:05 Pricing and Agent OS Setup
04:05 GaudiBench and Context Window
04:53 More Builds Mixed Results
07:11 Games and Visual Quality
07:59 Web UI and Web OS
09:08 Final Verdict and Leaderboards
10:19 Dont Chase Models
10:34 Agent OS Offer and Wrap Up
Full transcript
Sonic 5 versus GLM 5.2 the one-shot showdown who wins we're gonna walk through it today and side-by-side we'll be comparing how Sonic 5 compares to GLM 5.2 so let's get straight into this and the first thing that we're gonna start with is a crypt game like a dungeon crawler that we've created with both of these models so this is GLM 5.2 this is Sonic 5 which one wins let's compare them side-by-side we'll also compare the benchmarks in a second as well so if we have a look this is the output from GLM 5.2 and not bad not bad at all let's have a look at the output from Sonic 5 I'll actually say the graphics are smoother but the gameplay itself is not that great like there's literally nothing going on here it's super dark and there is no game to be played it's just you walking around a maze so not that great and I would say this is kind of the theme that I've seen across everything and we'll come on some more tests in a second in terms of how they perform so I also got another one here which is like you know build a ray caster maze so let's see how they perform this is the output from GLM 5.2 and if we have a look here you can see that is super buggy right it's really really buggy as you can see here whereas for example if we have a look at the output from Claude Sonic 5 actually does win in this particular contest so it is possible to get better outputs with Sonic 5 than GLM 5.2 bear in mind that Anthropix Sonic 5 is closed whereas GLM 5.2 is open source so that plays a big difference as well and also of course GLM 5.2 is a lot cheaper than Sonic 5 so let's see how they perform on the tests here if we compare them on cursor bench what Fable 5 max right at the top extra high so Fable 5 is absolutely dominating crushing here if we have a look at Sonic 5 it's way down the charts you can see here for example it's number 13 however GLM 5.2 is actually a lot lower so you can see it scores 61.2% for Sonic 5 on cursor bench whereas Sonic 5 sorry GLM 5.2 scores 54.6 so there is a big difference here in terms of how they perform side-by-side now you also might be wondering okay should you pick Opus 4.8 or should you pick Sonic 5 if you're using Claude, Claude Code etc so what we can see so far is the Opus 4.8 crushes Sonic 5 on the benchmarks as well so if you have a choice between them I'd go Opus 4.8 but bear in mind that Fable 5 is expected to drop within the next 24 hours that's been announced officially from Anthropic so you know if you had a choice I'd put Sonic 5 down here then we've got Opus 4.8 above that and then we have Fable 5 that's even better than all of them and also something to bear in mind as well you see in this tweet by Lisa one point is 1.2 times more expensive than Opus 4.8 and Sonic 5 is two times more expensive than GPT 5.5 actually five times more expensive than GLM 5.2 so bear in mind like if you're using the coding plans so you can get a coding plan with Claude or you can get a coding plan with GLM 5.2 if you're using the coding plans it's gonna be a lot cheaper with GLM 5.2 and also bear in mind like if you want like the full power of GLM 5.2 but you prefer using something like Claude Code we actually set up something called GLM Code inside the Agent OS system so that we can use the power of GLM 5.2's brain inside Claude Code and then we can see everything that we've created right here and it performs pretty well when we're doing that as well so you get like the agent harness of Claude Code but you can plug in GLM 5.2 and then you know either way whatever drops whatever comes out next you can get the most of these models by having great systems like this now let's have a look how it scores on GoldieBench so obviously Claude Sonic 5 just dropped a few hours ago it's a million token context window it's scoring pretty well on SWBenchVerified to be fair they've actually said it's their most agentic Sonic yet but I mean that should be the case anyway right I would expect that anyway like for example if you're using the older version of Sonic 5 are they gonna be as good as the new Sonic 5 no but at the same time how does that compare versus Opus 4.8 well Opus 4.8 is crushing it and that's that's the difference right here if you actually look on the benchmarks as well you can see the Opus 4.8 is pretty much outperforming on everything it's just that Sonic 5 is a big step up from Sonic 4.6 that's a big difference here so if you want to see some other stuff that we built with Sonic 5 let's have a look here so the outputs were kind of disappointing and you can see with this like diagram it's just a little bit boring the orbit didn't work the synth wave is okay but it's not doing anything special the one thing I would say as well is like I'll keep GLM 5.2 in the agent OS but I won't be using Sonic 5 and also bear in mind like one big difference between both of these models is that with GLM 5.2 you can actually use OR with Hermes agent or OpenClaw which means you can use them as the brain to power your agents so you can use them inside Claude code you can use GLM 5.2 with their CLI you can use it agentically inside Hermes or for example OpenClaw but you can't do that with Claude because they block OR login which means you have to pay for the API now we did do some other things as well so this was a game that we created called Dragon Realm from Sonic 5 which is it's not bad I mean it's a little bit basic in terms of the responses and everything else but it's not bad at all now if we compare that versus GLM 5.2 this is GLM 5.2's output I would say it's just slightly better slightly more interesting bear in mind like you've got more characters that you're gonna come across inside this game whereas if you look at the output from Sonic 5 there's not much really going on right when we look around there's no characters or anything like that it's just kind of like a basic landscape the same for example we created like this game as you can see which is quite fun easy to play with GLM 5.2 but then if we have a look at the output from Sonic 5 is you look at this and you think what is this I think we created something like this 12 months ago with with Manus or Genspark so you know that's a the common theme that I'm seeing right here but then you look at something like this this is created with Sonic 5 and it looks a lot more interesting and it's a lot better than if we for example compare this versus GLM 5.2's output which is just super boring on this so here's a mixed bag like you know Sonic 5 does hold its own against this but again like I think the Opus 4.8 would do even better both of them this was a game called Neon Drift kind of like a 3d racer game this is created with GLM 5.2 and then this is the one created with Sonic 5 which is still pretty cool I mean it's like a 3d game quite fun to play etc so here's another one this is a game kind of like an RPG open-world so we called it Dusk Wanderer and if we look at the landscape and the colors from GLM 5.2 it's a lot nicer it feels a lot better than for example if we compare that to Sonic 5 here look how basic this is nowhere near the same standard or level right and if you want to check these out for yourself you can check them out side-by-side on Goldie Bench here's another one so this was a website that we created it feels super basic with Sonic 5 like for web design and UI whereas if we compare that versus the output from GLM 5.2 it just feels a lot cleaner it's a lot nicer I mean it does use the generic AI colors the purple gradient but everything else apart from that is pretty good and then if we have a look here we've also got web OS so this is literally an operating system this is the version that we got from GLM 5.2 and if we create something here it seems to work perfectly it saves this application then we close it and if we open it back up again it should have our saved notes inside notes here so look yeah so it saved everything that we created and then for example we can open up the calculator app and it's ready to go and so that works pretty nicely this is GLM OS now let's check the same thing with Claude you see how the background just feels a little bit a little bit basic it's not smooth versus this one and then let's check the notes app see if it actually saves so we'll type some notes close that open it back up it does save it to be fair but it just doesn't feel quite as nice there's something about it that's missing in terms of the final finishing touches so overall if I had a choice between them I'm gonna go I mean I would stick genuinely with GLM 5.2 it's cheaper we can use it agentically with Hermes it's open source we get a lot of tokens on the plan and it consistently outperformed SONNET 5 on the benchmarks now if you look at the actual leader boards here on Gaudi bench you can see Fusion and Hermes mixture of agents are still right at the top GROK minimax doing pretty well as well and then you've got GLM 5.2 down here we closely followed by Opus 4.8 and then SONNET 5 right now if I had a choice between all these I'm probably gonna go with Fable 5 and we'll test out again within 24 hours once it comes back and it should be out back out to market the export controls just got re-released got dropped for that so that's gonna be pretty interesting and then you know whatever happens we're gonna plug it into our systems we're gonna make the most of it if you want to get like for example GLM plugged into Claude code as a setup we've got over here if you want to use GLM alongside Claude we've got that set up inside the agent operating system over here we have a memory system that links them all together and that's automatically organized by our AI agents and then the thing that I would say here is don't chase a model chase a system right so whatever happens with GLM 5.2, Fable 5, SONNET 5 doesn't really matter because we can make systems like this that are super anti-fragile and just a very flexible to them so if you want to get my full setup that's inside the agent OS in the AI Profitable Audit link in the comments description or go to AIprofitableaudit.com and inside the community you can ask questions I personally answer all the questions inside the community daily inside the classroom you get access to all of our best trainings we have a complete beginner to expert course over here and then if you want to get the agent OS we have that inside this section so you get the agent OS is updated daily you get video tutorial a full guide and the zip file to install it and then we add new daily tutorials based on what's just been released as you can see now also inside the calendar you can jump on weekly coaching course get help and support in real time you can also direct message me and ask me any questions if you want to and then inside the map you can connect with people locally near you who are also building with agent operating systems and everything so feel free to get it link in the comments description or just go to the AIprofitableaudit.com thanks for watching
More episodes