AI News Today
← All episodes
Episode 92 · July 27, 2026 · 13:45

Claude Opus 5 DESTROYS Fable 5?

Full transcript

Claude Opus 5 just dropped and I can tell you right now we've tested it across many different builds already, only just dropped about an hour and a half ago and it is an absolute banger. So I'm going to show you exactly how it works, how it performs, what the benchmarks are like and what we've built with it so far. So just to recap, this is the announcement of Claude Opus 5, designed to be basically like the same level as Fable 5 but it costs half as much. And you can see the full details here in terms of how it performs on the benchmarks.

The reason I wouldn't pay too much attention to this is just test it out yourself because it's always different when you actually test these things, especially when the model first drops. So that's why we've built so many stuff as you can see right here, we've built all this stuff and I'll show you how it performs in a second. But yeah, it's pretty good for agentic terminal coding, knowledge work, it's also pretty good for computer use but the main point here is like it costs a lot less. So for example here you can see Opus 5 vs Opus 4.8 vs Fable 5 and it basically achieves like similar levels of performance and even at higher levels of effort than any other model, it's cheaper if that makes sense.

So that's one of the highlights and also very good at computer use. So you can see the announcement right here, you can check out all the benchmarks, I'm not going to bore you with them because I think most people are kind of over just generic theoretical benchmarks. You're probably just wondering, okay, is it good or is it not good? So let's get straight into this and let's have a look.

So these are the builds, you can see here for example this driving game and we've compared it vs GPC 5.6 Sol over here, then we have Kimi K3 over here and we have Fable 5 over here. And actually even if you compare it like side by side vs Fable 5, on this particular build I'm not saying it's better than Fable 5 but I am saying like it's right up there and it's super nice in terms of the detail, what it can create and that sort of thing. Some people as well will watch this and be like, well, you know, why are you just using it to build games? Games just give you a good feel for like the look, the feel, how well it can predict physics, how well it can plan properly and also the level of detail and the vibe of stuff you'll get when you're building with this but so far pretty cool.

Let's have a look at the next one, so this is Dragon Flight as you can see right here and again like the graphics, the vibe, the details, the physics, everything is pretty much on point when it comes to this, it's holding it's own if not absolutely destroying some of the other options. So you can see how it performs and compares vs other models and I would say it's right up there. Let's have a look at the arcade one here, so this is kind of like a 3D snake game that it's created. As you can see, honestly this one is not like that interesting, not that fun to play, basically it's like a first person neon snake game, that's kind of what it looks like.

Not quite as good as the rest to be honest, not as fun to play. Let's have a look at the next one, so this one is cool, if we have a look here, this is pretty cool, kind of like just a fun game to play, 3D, controls are nice, easier to use and more fun than the other stuff that we've created here. So if we look for example at GPT 5.6, so it's good but it just feels like a little bit weird when you're playing, the colours and everything and the same for example if you have a look at the Fable 5 comparison here, the controls are just a little bit less nice and also when you're looking at this, the detail on the actual setup here from Claude Opus 5 is much nicer and then if you have a look for example at Kimi K3, Kimi K3's version is not bad but it's just a little bit boring. So let's have a look at the next one, so this is like a flying simulator game, again, holds its own if not better than all the other models, which is unbelievable when you think about it.

We'll be running like 50 different builds and giving it a full score later on Goldie Bench, seeing how it performs on the leaderboards but I'd honestly say you might be looking at one of our newest, most favourite models. I think the only things that are actually going to outperform it here are when you've got fusion panels, so for example if you have Fable 5 and Opus 5 working together, that would be a nice fusion to see how they perform and how you can get better outputs. The only thing I will say on this particular example is the controls are kind of weird, it feels a little bit weird when you're flying around there. Let's have a look, we've got the flight simulator, this one didn't seem to work at all actually, if I'm just testing the controls here, it doesn't seem to work, so it's not perfect compared to these, but bear in mind as well, it's just so new, with these builds these are all like one shots, it's just taken one shot to build these.

So if some of them are not perfect, it's okay, we can go back and forth, we can test it a bit more later, it's still building some stuff out as you can see, but really really good, really impressive so far. Now you might be wondering, okay how do you get the most out of Opus 5? So I call this the Opus 5 command engine we've called Opus 5, we've actually built it already into our agent operating system and I think that's the best way to get it, you can get it inside the Airprofit Boarding, link in the comments description or just go to theairprofitboarding.com, but yeah it's a fun, powerful model, it can build anything that you want as well. So it's going to be pretty good for computer use, it's better than Opus 4.8 which is already pretty decent at computer use, so this is only going to get better and the main thing that I would say here as well is that if you look at Fable 5, it's kind of limited because you only get it at 50% of your subscription on Claude, not ideal, whereas for example Opus 5 you can use, it's going to cost you less tokens on your subscription, but at the same time you can use it more and you can use it more than your 50% allowance on Fable 5 which is even better, so you just basically get the same model, it's still super powerful, it doesn't outperform Fable 5 on everything, but on a lot of the key benchmarks and the tests that we've already run, it's really really good.

Let me show you another example, so here's like a open world RPG game that we created, quality is nice, 3D graphics are nice as well, even the world, like the way it moves is pretty interesting as well, like you can kind of see it swaying side to side as well, graphics are nice, details are good, overall it looks really really cool. And here's another one, so it's like a 3D racing game, through an open world, and honestly it's like better than most of the big frontier models that we've tested side by side, so genuinely like Opus 5 can create awesome stuff. Here's another one, so this is like a 3D neon blaster game, and these are all just one shots, but it did a great job. So what are the key things to take away from this, so number one, coding, more than double Opus 4.8, so on frontier bench, Opus 5 scores over 2x the old Opus 4.8 at low cost.

Number two, reasoning, three times the next best model on ARC, AGI 3, a benchmark for genuinely novel problem solving, Opus 5 scores about three times the next model, which is pretty big. Real work, it tops the automation tests, so on Zapier's automation bench, it passes roughly 1.5x the rate of the next competitor, and hits 100% on end-to-end churn prevention workflow. Science actually had a massive jump on that as well, and then like actual work work, like office work, it did pretty well too, and additionally it's safer as well. One thing to note is like this will automatically root, just like Fable 5 does, automatically to Opus 4.8 sometimes, so just be aware of that, like that's going to be kind of annoying.

It did that with Fable 5, it will do that with Opus 5 from time to time, they actually put that inside the release notes, like right at the bottom of the official page. But overall, Frontier Bench, Cursor Bench, all sorts of benchmarks are doing very well, on my own tests it's doing very well as well. Now we've already put Claude inside our agentic operating system, so what you can see for example is that we can use the chat here, and this is using the CLI so we don't need to pay for the API, and then everything that we build is like plugged into our workspace over here as well, so we can like preview it, we can have a look at what we've built, and then additionally we have these artifacts, and we have ultra code built in as well, so it's like all the power of Claude, but in separate workflows where you can just access them in one click, and then also because this has literally just come out today, like less than two hours ago, the great thing about this is that we can actually plug it straight into a memory system like Obsidian, which we've got inside the AgentOS as well, and then it's got instant memory, instant context straight away, and additionally it will just work side by side with Hermes, with Antigravity, with Codex, it will be plugged straight into our group chat over here, and that's the beautiful thing about having a system rather than relying on the model, so whatever comes next, whatever came before, doesn't really matter, we've got a great system that improves every single day, and that's the way that I'd recommend using any of these workflows for yourself as well. Another one for example is we do a lot of SEO with Claude inside our system as you can see right here, and using Claude with the AgentOS for SEO is really really powerful, it helps a lot of our websites, and we can now just plug Claude Opus 5 straight into there, which is what we've done, and so when you've got a command engine like that, it just gets better and better every time you use a different model like this.

You also might wonder, does this use up a lot of tokens, but the great thing is that number one, you can switch this for free APIs, like for example we have OmniRoot plugged into this if we ever run out of API tokens, sorry, if we ever run out of tokens on our subscription for Claude 5, but additionally we can plug in free APIs, we can plug in local models, it doesn't really matter, like I never really worry about tokens, it's not a big deal if you've got an AgentOS, because we've got the models that are built around making sure it's super efficient in the way that we use it as well. But yeah, it can create some amazing stuff, holds its own, you don't need to worry about for example the cost of Fable 5 anymore, you can plug in your CLI into the AgentOS, it performs better on many benchmarks versus Fable 5, and it's also got a fast gear as well, so it can run two and a half times faster with fast mode. That does use up more tokens and it's more expensive, but that is an option as well. And the great thing about having an Opus 5 command engine like this is we have the core, which is Opus 5 itself, which is near Fable 5 Intelligence, but then we also have the console, which is the AgentOS dashboard, you've got Claude, you've got Hermes, you've got OpenCLR and Codex in one place.

Then you have the memory layer, so that's layer number three inside the system. So that's your obsidian vault with every session saved and Opus 5 understands everything about your business and doesn't just work in a vacuum, but works across all your different AI agents. And you have a fleet of AI agents working side by side as well. So you can run loads of jobs in parallel using, for example, Hermes, Claude, we have the pipeline that actually takes you from idea to app and implementation with Opus 5 and other AI agents.

Plus we have Paperclip, so you could have, for example, Opus 5 from Claude working alongside Hermes, working alongside AntiGravity, and then you can just have them inside a nice team like you see right here to orchestrate them all together. And then we have the proof, you know, we've tested this across many different builds and it is creating some awesome stuff as you've seen today. So you might say, OK, well, you know, Opus 5 is not as good as Fable 5, but it lands within 0.5% of Fable 5 on CursorBench. The only difference is it's cheaper and then you get more tokens with it on your Claude subscription.

You might also say, I'll wait for the benchmarks to just be fully completed before switching. Honestly, from my own tests, I wouldn't even wait at this point. Like it's just a no brainer to switch. And additionally, you might think, well, the model is remote, so I just need the best one, I need Opus 5, I need Fable 5.

Actually, you don't. So even if you don't use Opus 5, the system around the model is remote. So these models come and go every single month at this point, sometimes every single week. The main thing that you need is a system like this to manage it all.

Otherwise, you just get left behind. And you might also think, OK, well, doesn't the AgentOS with Claude Opus 5 make things like a bit complicated or a bit technical? Absolutely not. So, for example, Andrea was saying like how having a mission control operating system, she actually got ours from the Air Profitable, link in the comments description or go to the airprofitable.com, has been an absolute game changer.

You can see Hakim absolutely loving it, Kurt as well. He said the AgentOS has changed everything for him. So loads of people are winning with this stuff. We've got over 200 pages of testimonials of wins, people growing, learning and scaling with AI automation.

So you don't need to be technical. I'm not technical. I don't understand code, but I can create amazing stuff with this and so can you. That's the whole point here.

So if you want the full Opus 5 command engine built for you, very powerful model, creates awesome stuff, but you need a good system around this, then you could create this yourself. But if you want to get our full system, you can get the full AgentOS zip file, the Obsidian memory system, coaching calls where we set up your command engine together, step by step, a room of 3,900 founders running this exact stack and also the prompts, ESPs, daily tutorials and everything else inside the Air Profitable. We've also got a Claude Opus 5 masterclass inside there. If you go inside the classroom here, if you want to get our AgentOS, you can get that like so and you get the video tutorial, the last update date and the zip file to install it.

Plus we add new stuff all the time. Inside the community, you can ask questions, get help and support in real time. And I personally answer them as well. And there's always people online ready to help you.

And like people are just helping each other out. People are just absolutely loving it, enjoying the journey, sharing with each other, making great friends. And that's the great thing about it. And then also inside the map, you can meet people in your local area who are building with Claude Opus 5 and other models.

Plus inside the calendar, you can jump a weekly coaching course. We have four weekly coaching course. You can jump on a call, share your screen, ask questions, ask our coach anything you want as well. And then inside the classroom too, if you're a complete beginner, you can go from beginner to expert over here.

If you want a full step by step courses, you can get that here. These are like one hour courses that guide you through everything you need to know. And then if you just want like the quick short tutorials, you've got that over here and these are newly updated all the time as well. So feel free to get that.

I hope to connect with you inside the Air Profit Boarding. Plus you can direct message me there anytime as well. Thanks for watching. Bye bye.

More episodes

Browse all episodes →