Transcription
Claude Opus 5 is here, a day later than it was supposed to come out, but I'm not even going to bury the lead here. It is incredible. It is so much more powerful than I thought it was going to be. And the best part is you don't need to pay for Anthropic's most expensive plan to access it. If you pay for Claude Pro, you already have access to Opus 5. So, let's not waste any time here. Let's talk about the new Opus 5.
Opus 5 is now Anthropic's strongest Opus model. It's available inside of Claude on Pro, Max, team, and enterprise plans. If you have Claude Max, then when you come over to a new Claude chat, you will see that Opus 5 is the new default. If you have a Pro plan, you can choose Opus 5. It just won't be selected by default. You can also use it through the API or inside of Claude Co-work or Claude Code.
Now, Anthropic already has a ridiculously powerful model called Fable 5. I've made plenty of videos about that already because I felt it was the first AI model where I honestly felt like it could take over parts of my work without me constantly needing to babysit everything. But Fable has two problems. First, it is expensive. And second, a lot of people, if you have the Pro plan, you're not going to have normal access to it. You'll have to pay extra to use Fable 5. Opus 5 is designed to bring you very close to that level of intelligence at half the price. And unlike Fable, Opus 5 is included with regular Claude Pro plans. So you will have access to it no matter which Claude plan you're on. And that makes this a much bigger update for normal Claude users.
Now, I normally don't spend much time on AI benchmarks. What matters is whether it can actually complete useful work. But some of these results are difficult to ignore. So right here, if we look at how well it handles knowledge work, you can see that it outperforms every other state-of-the-art model, including Fable 5. If we also look at how well it handles novel problem solving, it is doing very well compared to the prior Opus model, Opus 4.8, and even better than GPT 5.6 Soul. It's leading when it comes to agentic search for computer use and for business workflows. There are still some weaknesses, like it's not as strong as Fable when it comes to legal work or health work. But for the type of work that I know a lot of you do, which is knowledge work, this really seems like this may end up being the best model for the job, even better than Fable.
Zapier also publishes this automation benchmark, which basically tests how well an AI agent can perform end-to-end workflows on its own. And Claude Opus kind of cleaned up here. They, every single version of Claude Opus, in terms of the thinking level, has outperformed every other model, again, including GPT 5.6 Soul, even on the lowest effort. And because of that, I think that Opus 5 is going to be an incredible model to use inside of Claude Co-work, even better than Fable. Not only because we don't have to worry so much about burning through our usage when we use Opus instead of Fable, but just because it seems to, at least according to the benchmarks, be better for knowledge work than Fable 5.
I was able to do a few quick tests with it and I saw some interesting results. So, first, I had it create for me this landing page. Now, I don't normally use Claude for work like this, but I thought visually it would be a good representation of the model. So, this one was made by Opus 5. I basically asked it to create this self-contained HTML file for a product called Houseion, which is a noise-canceling desk lamp. And I thought that the page that I came out with looked pretty good for, for Claude. I also produced the same one using 4.8. And immediately I noticed a big difference with it, like this looks much worse than Opus 5. And then finally, I created it with Fable. And I'll say I think that Fable is maybe slightly better. Like this part right here looks a little bit better than what we saw with the version made by Opus 5, but not too much better. And the one made by Opus 5 was faster.
I also asked Fable 5 and Opus 5 and Opus 4.8 and Sonnet to put together the script for this YouTube video. Just gave it the idea and had it go out and do the research and put together the script. That's normally not what I do. I don't normally fully script out my YouTube videos, or at least have AI do it for me, but it's a good experiment for me because I understand what makes a good video for yourself. What I really recommend that you do is create your own benchmarks. What are the tasks that you normally give to AI, AI tools, and treat those as a benchmark. So when these new models come out, you can throw a similar task to the new model and see how it compares to the results that you're used to getting.
Anyway, with this little test that I did, I saw that Opus 5, the script it put together for me, was about on par with what I saw Fable producing. Neither of them, quite honestly, are that amazing. I definitely wouldn't use them as they are when recording a video, but they're better than anything else I've tried out there from Anthropic. And more importantly, the results between Fable 5 and Opus 5 are on par with each other, but of course, Opus 5 is a lot cheaper.
And finally, I asked both Fable 5 and Opus 5 inside of Claude Design to put together this motion graphic for me. And you can see these look pretty much exactly the same, which is kind of the point. Exactly the same results, but for half the cost. And again, Opus was much faster than Fable 5 by several minutes.
I want to be clear, of course, that this is a launch day first look. I have not had Opus 5 long enough to pretend that I've tested every corner of it. We've also seen plenty of new AI models look incredible during carefully selected demonstrations and then behave very differently once you normally start to use them. So all these examples are just quick demos of its capabilities. I really want to test this a lot more with my workflows and see how it performs. But in my initial impressions of it, I am very impressed. It feels to me as good as Fable for knowledge work, but actually usable as a default model, which is very impressive and is something that I'm going to be using a lot more again as my default model because why not? Super intelligent at a price that's actually affordable. It's pretty nice.
But test it out for yourself. Throw at it some of the work that you normally use with Claude or ChatGBT and see how it performs against the model that you're already used to using. Then share your results in the comment section down below so we can all get a grasp on how good the new Opus 5 is.
Next, if you want my full breakdown on how to use the new Claude Co-work and how I use it with my own workflows, click on this video right here. Claude Co went through some major revisions over the past few weeks, and this is probably still the most up-to-date video right now on how to use it. So, make sure to click on this, and I'll see you over there in 1 second. Bye for now.