Claude Sonnet 5 is HERE!

So today we have the release of Claude Sonnet 5, apparently the most agentic model from Claude and Anthropic yet and apparently you can see the announcement here, just dropped a few hours ago, it can make plans, use tools like browser terminals, run autonomously at a level just a few months ago required larger and more expensive models. We'll come on to this in a minute, I'm not going to hype it up here because you'll see from my test. I'm just going to tell you the honest truth. So here you can see you've got Sonnet 5, you've got Sonnet 4.6, so it is a step up from Sonnet 4.6.

If you actually look, like Sonnet 5, nowhere near the same level as Opus 4.8. So if we look at agentic coding, 63% versus 69% for Opus 4.8. I mean literally on every single benchmark here, Opus 4.8 is crushing it. Now if you were comparing 4.6 from Sonnet versus 5, then of course it's a big step up.

But I think for most people, you're probably still going to stick to Opus 4.8 for building stuff, especially at this point. Now we can expect Fable 5 to drop very soon. Anthropic just announced that they're going to be releasing and giving people access back again within 24 hours for Fable 5. So this kind of just feels like a weird release where I don't know if many people care about it.

I mean if you're watching this, you use Sonnet 4.6 already, are you not just sticking to Opus 4.8? Are you not going to just switch to Fable 5? For me personally, I just don't know if this is really a big update or not. Now if we actually have a look at what we've created so far on GoldieBench with Claude Sonnet, let's take a look right here.

So if we pull up the benchmarks over here, he did create some pretty cool stuff and I'll show you what we've built, how it works, etc, how it performs. So we can gather some examples side by side over here. Here's a couple of examples. So this was like a raycaster maze.

I mean it can create pretty nice stuff, right? Like this is pretty cool. The graphics, pretty nice. The colors, pretty nice.

I mean it works perfectly as a game. For some reason the controls are a bit backwards when I test it out personally. But apart from that, we're all good here and it's working nice. Now if we have a look at this one, this is kind of like a little test.

The orbit test for a galaxy didn't work at all. So some things had broken on it completely and then this was kind of like a cool synthwave background as well. Then we actually created a crypt game as you can see right here and it looks pretty nice. Like I mean it's again, it's smooth, easy to use, etc.

Not bad at all. But when we actually look at it on the benchmarks, you'll see how it compares in a second. You also might be wondering, okay how does it compare versus GLM 5.2? I'll do a deeper tutorial on this later.

But basically if we have a look side by side, this is the maze example from Sonic 5. Pretty nice. I do think it's a big step up versus what GLM 5.2 created over here which is, you know, using the same prompt but it's just a little bit buggy, not quite smooth. You can see, I mean a little bit buggy, I would say very buggy at this point.

So when we've compared it on tests like that, it didn't do so well. But then there were other things where, for example, the orbit didn't work at all with Sonic 5. But if we have a look at GLM 5.2's version, it actually works really nicely and it can complete the task. So there were some tasks where Sonic 5 actually totally failed and then GLM 5.2 crushed it.

But we'll come on to more stuff like that later in a different tutorial. Now we can also see here, for example, side by side, we've got two outputs from Sonic 5 side by side for this game as you can see. You can see it didn't do so well in terms of public reception. So, for example, you can see this tweet from Lisa and he said, Sonic 5 goes straight into the garbage bin.

1.2x more expensive than Opus, 4.8 max, two times more expensive than GPD 5.5x high, five times more expensive than GLM 5.2. I mean, it's pretty wild when you put it like that. I can understand why a lot of people are quite negative about it because basically Opus 4.8 is outperforming Claude Sonic 5 on many benchmarks as you saw earlier. But Sonic 5 is more expensive than Opus 4.8.

So Opus 4.8 is better, but Sonic 5 is more expensive. How does that make sense? And again, I think with Fable 5 coming out soon, this is going to be one of those things where people don't even talk about Sonic 5 anymore because it's just, it's been received so badly. You can see another tweet here from Bridgemind.

So he said, Anthropic seriously messed up with Claude Sonic 5. The token efficiency is so bad it actually ends up being more expensive than Opus 4.8. What are they doing? So it's been quite a brutal day for Anthropic.

Here's another one from Bridgemind who was saying, you know, Claude Sonic 5 looks terrible on CursorBench. The whole point of using Sonic is that it's faster and cheaper. But if we actually look at the leaderboards here, we can see that Sonic 5 is way underperforming Opus 4.8. So if you're going to ask me like, okay, what is the best API to stick to?

What model would you use with Claude Code or Claude Desktop, whatever, just stick with Opus 4.8. Like there's no point using Sonic 5 from what I can see right here, especially when it's scoring so low on GoldieBench. I mean, let's compare it to a couple of other things. So we've got Claude Sonic Fugu Ultra.

So Fugu Ultra is the new model from Japan. So this is the output for the Crypt game from Fugu Ultra. And basically what it does, it fuses two models together and you can see like the output is pretty nice from that. Now, if we have a look at the outputs from Sonic 5, you can see like the colors are a lot darker.

There's not as much going on. It's not as interesting, etc. So even like the models where you fuse older Frontier models together are way outperforming Sonic 5 by quite a long way. Like this is just a much more interesting response.

Let's have a look at another example. So we've got liquid in a bowl here, right? So what should be happening here is that you have like liquid in a bowl. And here's an example of what it looks like from Sakana Fugu.

Now, if we compare that to Sonic 5, it just doesn't look as good, right? And that's a huge gap in quality. Again, if we compare the Orbit outputs, Sonic 5 totally failed. Sakana Fugu crushed it.

Now we're going to do more tests and it just dropped out a few hours ago. But from what I can see, I just, I wouldn't use Sonic 5. I don't see the use case. It might be great at writing potentially.

That might be something, but apart from that, yeah, there's no need to switch or anything like that. And you're probably asking, why would you release a worse model that's more expensive? I don't know if like, for example, Anthropic just felt the need to like bring out something new because there hadn't been a new version of Sonic for a while. I'm not sure what they were doing here and whether they were just disappointed themselves with the outputs.

But from everything that I can see, I just, you know, I wouldn't, I wouldn't use Sonic 5. What I do like, what I do actually appreciate is that they were so honest on the benchmarks, because there's probably a lot of companies who would try and make Sonic 5 look better than it is. But from what I can see, like the benchmarks were ruthlessly honest. The Opus 4.8 is a lot better.

Either way, whatever happens, you know, when there's a model launch like this, focus on the system. Don't worry about the models. Fable 5 will come out pretty soon as well. And so, you know, I wouldn't worry about like, what's the hot model of the week?

You want to have a system where you can use something like the agent operating system and it doesn't matter what comes out next. So for example, for us, if Sonic 5 was actually good, we would plug it into our agent operating system. It's not, so we'll keep Opus 4.8 instead. And that's the whole point here is like, whatever happens, you want to have systems ready to go that are flexible, no matter what comes out next.

I mean, for example, something else that we did is we plugged in GLM 5.2 into Claude's code so that we could get the best agent harness with one of the cheapest models. And then we can combine the best of both. And we've got like a workspace where we can track everything. We've got a memory system plugged into all of this.

We've got Hermes agent ready to go and Claude CLI and everything else. And then we've got a team of agents inside our group mastermind and everything else. So the point here is like, focus on the system and how many great systems. And then if good models come out, you can swap them in and out.

If they don't like this, you're not disappointed because you still got great systems and it doesn't matter what comes out next. So if you want to get our agent operating system, it's inside the AI Profitable Boarding. This is our community that helps you save time and grow and scale with AI automation. Inside the community, you can ask questions, get help and support whenever you want.

Inside the classroom, we get access to all of our best trainings. If you want to get our agent OS system, it's over here with a video tutorial, a step-by-step guide and the zip file to download and install. And then you can see there are new daily tutorials right here. Also, if you want to jump on weekly coaching calls, we've got four coaching calls per week where you can share your screen, ask questions, get help and support in real time.

And inside the map, you can actually meet people in your local area who are building with Claude and agent operating systems and stuff like that. But overall, SONIC 5, not great, wouldn't switch, continues in Opus 4.0 and let's hope Fable 5 comes out soon. Thanks for watching.