Claude Opus 5 is INSANE!

**Julian Goldie** (0:00)
Claude, Opus 5 has just dropped, my friends, a thoughtful, imperative model, according to Anthropic, at half the price of Fable 5 So it comes close to Frontier Intelligence, at half the price of Fable 5 And you can see how it performs on the benchmarks today. We're already running it through Goldiebench, so we'll test it out and see how it performs across 50 different tasks that will be coming to you soon. And if we have a look at Opus 5 here, this is Fable 5 versus Opus 5, versus Opus 4.0, versus GPT 5.6 Sol. And what it really excels at is agentic terminal coding. So you can see Frontier Bench, and it's absolutely crushing at 43.3% versus Fable 5 at 33.7%. And GPT 5.6 Sol at 34.4%. It's better at knowledge work, better at problem solving, better at agentic search. So those are four things that is better at outperforming all of the other models right now. When it comes to multidisciplinary reasoning, it's kind of mixed on this one. So it's excellent. And it's a lot better at computer use versus these other options. Maybe something coming from or desktop soon on that, I reckon, related to voice, if ChatGPT is anything to go by. Then we have, for example, agentic coding and business workflows performing really well. You can see here Fable 5 is still beaten at legal and health and then biology of all things, Opus 5 is excelling out as well. Bear in mind, like, the difference with, for example, Fable 5 is that it's only including 50% of your chat, of your Claude subscription. Whereas, for example, if you are using Claude Opus 5, you can use 100% of your subscription usage of Anthropic to get the most out of it. So you get frontier level technology in a way where it is not limited on your current plan. Let's have a look at how it performs versus other agents. So the thing to note here is like, if you're using the API, then the cost per task is cheaper, right? So genetic computer use performance, for example, by effort level. And you can see here, the Opus is cheaper even when the effort level goes up. So for example, this is Fable 5 in orange, and this is Opus 4.8. So Opus 4.8 is more expensive than Opus 5, even when you raise the reasoning levels. And then Fable 5 is obviously way more expensive than both of those. GPT 5.6 Sol, when you raise the reasoning as well, is still slightly higher. And it's the same, for example, for business workflows by effort, multidisciplinary reasoning, and also agentic coding. Also on ARC, AGI 3, in a valuation where AI models will solve novel problems, Opus 5's score is three times as high as the next best model. So we can see here how well it's performing.
Opus 4.8, all the way down here. So, I mean, the main thing to note here is like, you're basically getting Fable 5 level intelligence, according to this, you always want to test it out yourself. You're basically getting Fable 5 level intelligence on a model that is cheaper, or you can use with your full subscription. And if you look at the gap between Opus 4.8 and Opus 5, it's absolutely huge on most of the benchmarks here. They've also said that Opus 5 is stronger than Opus 4.8 on cybersecurity tests, but it remains substantially behind Mythos 5 at developing exploits. So it's safeguards are designed to allow developers to identify and fix software vulnerabilities whilst blocking high risk uses.
And it's available today. So for example, if you go over to Claude.AI, it's available right there. And then also we've already plugged it into our agent operating system. So you can see here that we've got Claude Opus 5 running directly, and then everything that we build plugs into the workspace. Plus it already goes straight into our memory system so that we can get the most out of this. Now, bear in mind, this just dropped like literally 50 minutes ago as of recording this. So it's straight hot off the press. I haven't had much time to test it yet. But let's have a look through the announcement here and see what we've got. Bear in mind that really there is a lot of pressure on Anthropic right now, because number one, they've messed people around a lot with Fable 5 Then you got GPT 5.6, which is absolutely crushing it and a lot of people saying they prefer GPT 5.6 to anything else out right there. GROK 4.5 is getting better all the time. And then you've got two models, QWEN 3.8 Max from China, which is right up there at frontier level with something like Fable 5 We've tested out on Goldie Bench. And you also have Chinese models like Kimi K3, open source, fully open source, but super, super powerful, like right up there with Fable 5 as well. So they have to drop something that's more affordable and more available, just like Fable 5