I Tried Opus 4.8...

Hinzugefügt: 2026-05-30

[00:00:00]Opus 4.8 is here. I put it through a proper test where you can visually see the difference compared to previous versions. According to the benchmark results, Oppus 4.8 is beating GPT 5.5 in almost every category except the terminal coding. And if you look at the jump from Opus 4.7 to OPUS 4.8, it looks pretty impressive. But benchmarks tell only part of the story. I wanted to see how it performs in the real world. So in the same moment when it was released, I gave it a proper test and I did it in my four elements project and this is a perfect test because here you can visually see how much the model changed compared to previous versions. So here we can see air element created with Opus 4.7 and I gave it a prompt. I'm always giving it exactly the same prompt and I'm telling it not to hallucinate and to copy from other models. I have a system.

[00:00:59]How am I doing that? And here we have Opus 4.8 and we can see obviously the visual difference between these two models. If we check here OPUS 4.7 this curtain thing, it is not moving much and it is not looking that good. But Oppus 4.8 8 looks almost like PlayStation one graphics and it even added like this plant or something like this. So it was thinking a little bit more than Opus 4.7 and after that I tried Opus 4.8 Max and I was really surprised. So it created completely some different thing. It added here three curtains somehow. It even forgot about the windows. Uh, I'm not sure how that happened. So, there should be a window right here for the air element. But it didn't add it. It just added these three curtains with different colors. And this one now, I can say this is some PlayStation 2 if we are comparing it like that. So, this is really impressive change from Opus 4.7 to 4.8 and then to 4.8 Max. I think this is crazy. And you can find this project on forelements.dev.

[00:02:18]I'll leave you the link in the description as well. And you can see how it compares like with different elements and also with different models and how everything looks. So you can see like some difference visually between different models. Also, I'd like to point out that the price is exactly the same like for Opus 4.7. They're saying here available today at the same price.

[00:02:41]And they also have a fast mode that is available for 4.8 8 and it's the same model at roughly 2.5 the speed and the impressive thing is that they made it three times cheaper than before. I actually tried it out and it's really working. It is cheaper and fast mode might become my new default mode when using Opus 4.8. And if you check here the claude code to be honest I'm not using cloud code that much. I was using it yesterday and it's a mess. I was trying to find like for three minutes where can I push to main and here if you go you can choose OPUS 4.8 and you can enable here fast mode and I think that high plus fast mode is probably the new sweet spot at least for me. If you're really hardcore you can go max plus fast mode if you're doing some big refactorings or something like that. So try it out. Tell me how it goes in fast mode. I'm really interested to hear your opinion in the comments and also the newest feature of Anthropic and that's the dynamic workflows. So for the hardest tasks, Claude makes a plan, runs hundreds of parallel sub agents and verifies its work before reporting back.

[00:04:00]So think of migration touching hundreds of files. And we have a perfect example on their blog right here. So to be honest, I'm not exactly sure for who they created this dynamic workflows. So this is not for one ordinary dev that is working some open-source stuff. This is for some cracked dev who is refactoring many things and we can see it right here in the example. So rewriting bun with the dynamic workflows. An example of what dynamic workflows can unlock at scale is the recent rewrite of bun. So Jared Sumner used dynamic workflows to port bun from zig to rust with 99.8% of the existing test suite passing. So roughly 750,000 lines of rust and 11 days from first commit to merge. Can you imagine that? Someone rewrote entire bun in just 11 days and that almost 100% of tests are actually passing. So I think that this is really crazy. I cannot imagine. I would really love to know like the price of this refactoring, how much this one costs and how much Jared spent basically tokens on this. I think that would be really interesting piece of information that they are not giving us right here in this blog. I actually asked AI to give me some rough estimate and it estimated from $50,000 to half million. So that's a big difference, but I'd really like to know how many tokens Jared spent. Overall, I would say that Opus 4.8 8 doesn't look like it is smarter than Opus 4.7, but it has like they're saying here sharper judgment. I mean, what does that mean? I'm not sure, but I think that it's trying to get some GPT 5.5 kind of thinking to be like more honest about its own progress like they're saying here. So, it's not like always just approving and always just nodding head whenever you say something to Opus. And I think that's good. I think it's good that they're working on that kind of upgrade and not only like to make it better, smarter. And other than that, it is much faster now. Like we are getting results much faster, especially with the new fast mode. And especially that new fast mode is now three times cheaper than before. So, we can even put fast mode to be our default. I think this is a good improvement on Opus. I still think that GPT 5.5 is in some terms better than Opus 4.8. And I'm still giving more trust in GPT 5.5 to create my my plans versus Opus 4.8 to create my plans. But for the execution, I think that Claude is the king. So, tell me your opinion in the comments. I'd really love to hear what do you think with Opus 4.8 versus GPT 5.5. And for more content like this, join the mighty horde.

[00:07:19]Subscribe.