Claude Opus 4.7 is AMAZING! Full Breakdown + Testing Results

Oh, this is a good one. Claude Opus 4.7 just dropped. It's Anthropic's new flagship model. While it is available on all plans, including the free plan, the usage is extremely limited. We'll talk more about that later. And there's a bunch of new capabilities interesting to anybody. And I'm here to talk about them because we tried all sorts of things with this model. But one of the things that stood out was its ability to process images 3 times the resolution. So for example, I just went to my YouTube studio, zoomed all all the way out. And this is all the data for 2025. Now, I can't even read this looking at my screen. I took a screenshot, went into Claude, picked Opus 4.7 that is now available here. And I said, extract, analyze every bit of information you can. And it just did it. Whereas when I tried the same thing in the previous model, Opus 4.6, it started scrambling and using various tools. It used 4 commands. Look at those commands. It tried to write code and use the code to extract the details because the model itself couldn't do it. And then it actually got stuck. I got an error message saying, hey, I ran out of tool uses, so I had to say continue, and then it eventually did it. In other words, this new model can take images 3 times the resolution without cropping in, which kind of means you can give it any sort of image and it'll see every little detail. But it doesn't stop there. This model looks impressive from the outside too. Here, I pulled together the benchmarks for some of the top models, if you happen to care about these, and it's a significant jump on, well, mainly coding performance, to be honest. What we care about here is real-world performance. Performance. And I looked across the internet, tried a bunch of our benchmark prompts, and hmm, mixed feelings. One, it is legitimately better at a few things they claim. This image recognition is incredible. I think its ability to design, its taste improved. I'll show you examples here in a second. But also on the flip side, this model is way more expensive to run. And if you're paying for a subscription, thinking to yourself, oh, Igor, I don't care about the API cost, I don't use it like that. Even on your subscription, you're going to run out of tasks quickly. Matter of fact, us benchmarking this, Daniel, who's been running test prompts with me all morning, he ran out of usage. He couldn't even get through all the test prompts and ran out of usage within less than an hour. And if you are using the API, you can expect to pay 35% more than before. And it comes in a sneaky form because if you look at the pricing of the API of this model, it's the same as Opus 4.6. There's no difference. But they changed the tokenizer underneath. What that means is they changed the way they calculate for how many tokens they'll be billing you, which at the end of the day just means that you're going to be paying 35% more for Opus 4.7 than 4.6 if the usage remains the same. And that increasing cost also translates to your plan on Claude.ai with all the subscriptions. I should add, now this is a sneaky move because just a few days ago Anthropic added usage-based pricing into their subscription plans. So if you hit that limit, before they used to flip the switch and your account was just blocked for a while. Now they'll pop up this little screen, which gives you the option of, well, you you can either upgrade your plan, that used to be present before, but you could also buy extra usage. So you could just spend $50 to keep working now. And then they release a better model that is really token hungry. Smart move. I mean, you can always switch back to the older models, right? This is hard to hate on because if you're in Claude, hey, if you want to use the older inferior models or something like Sonnet or Haiku, where you're going to get like more than 100 times the usage, you can always do that. Even Opus 4.6 is still available if 4.7 is too token hungry for you. Less intelligence, less cost. More usage. It's up to you. There's this Reddit comment that I really like, which says, "Opus 4.0 is so strong, it can now create a beautiful landing page, and it only uses 95% of weekly usage to do so." It's a bit exaggerated, but he has a point. Now talking about beautiful landing pages, let's get into this. Is this better than the ones before? Well, a lot of that is always subjective, but that's why I'm here. I'm gonna give you my subjective take on this. Yeah, yeah, it is way better. Matter of fact, I told you Daniel helped me research this video, and Daniel is incredible at researching, forming opinions, but he's notoriously non-technical. Whenever it comes to anything coding related or website building related, he's just like, okay, that's not my territory. You know, I know, I know what I'm good at. This is not it. Now here's an example that he came back to me with today in the morning. He's like, Igor, I built a tool that we might be using on a weekly basis from here on out. And I really like perked up because I was like, Daniel, tool, both of those in one sentence. This doesn't make sense. And then he sent me the file and look at this, steal the signal, ship the short. So you can put in YouTube videos. And then if you give it your Anthropic API key, it will turn those YouTube videos into short scripts and other content for social media for you. Now that's not revolutionary in itself, but few things. First of all, look at this design. I don't know about you, to me, this is a 10. It's gorgeous. The background, the color usage, the spacing, the fonts, the way it framed up this box with little edges, like for simple little app like this. It's impressive, subjectively better than other models. It's just me. Secondly, he one-shotted this, meaning he said one prompt with the idea and he built a fully working application. Shit. Look at that, we put in an API, a video we uploaded last week about Claude Mythos that is still not out. It created 5 hooks that it ranked by what it thinks is best. Created a thumbnail mockup, which is not even using an image generator, so this is meh. And of course gives you a script. All of it beautiful, functional. Guys, my point is this. If Daniel can create this app, so can you. So if you haven't tried this before, it's probably time to do it. So now look, I want to say one thing. That's a lot of information, a lot of different areas. And if that feels like a lot, it's because it is. We're living through a technological revolution, right? And it's impossible for an individual to keep up with all of this. And I've been thinking about this for years. And you know that the mission of the channel is to help people on this journey from going to confused and overwhelmed and barely holding on, or maybe using the tools and getting decent results. To a whole new level. And we define a clear map on how to get there and what the progression looks like within AI to achieve fluency, to achieve this final agentic level. And no, the answer is not just go in and try your best. From my experience, there's a clear progression that you need to go through to use agents properly and productively. And to help you with exactly that, we're putting on the AI Advantage Summit next week, April 23rd through 25th. It's a free day, completely free summit that helps you make sense of AI and shows you very practically how to use all of this stuff for yourself. Now, I want to be very clear. If you're already racking up thousands of dollars in API bills and using OpenClaw and vibecoding applications, building a SaaS company, this might not be for you. But for the other 80% of the people where AI is not their full-time thing, this summit is exactly for you. So sign up from the link in the description. It's completely free. And now let's look at what else is new with Opus 4.7. All right, a few more examples. So here's a fun one I always like to do. When it creates SVGs, it kind of shows you how motivated it is to work, how detail-oriented it is. Let's do one of a chameleon sitting quietly on a branch. And let's make the eyes follow the user's cursor. Oh, and there he is. It's kind of fun. Additionally, ran the typical benchmark that you might know from the channel. I always do this. A Death Star over LA. This is the result in Opus 4.6. It's not bad. It's, you know, a Death Star. This is LA, I guess. Some of the windows are outside of the buildings. There's a random oval shape here. It's okay. Opus 4.7 though, better. Not perfect, but better. There's mountains, I love the shading, there's no random shapes floating anywhere. I guess some of the windows are outside of where they're supposed to be, but it's just better. For comparison, this is GPT's best model, which is also quite good, but I think Opus 4.7 subjectively is better than both of these. Here's another fun one that you actually might want to try at home. So the benchmarks report that it's better at internet search than before. It's not state-of-the-art in a benchmark, that gold medal would go to Gemini 3.1 Pro, but you could kind of just give it a shot and say something like, do intensive research on you know, insert your name, and give me a massive report on everything you find. What I found is that this report on me was more comprehensive than anything I've ever gotten from a simple LLM web interface, you could call it AI assistant, before. Just try it for yourself. It's really a hardworking model. And apparently it's better at searching the web. So go give it a shot. But don't be surprised, a few prompts like this eat up your entire usage, and you'll have to wait or buy extra credits. And talking about the development side of things, well, there's a bunch of changes, but there's mixed reviews across the internet. And I think a lot of the negativity that is surfacing now is due to the price increase. I mean, people just get mad if they know, oh my God, the new flagship thing is going to cost 35% more than I paid up until now. And then they're going to nitpick all sorts of problems with this. So, you know, when you see those opinions, filter them through, ooh, how much of that is just the financial distress that these people will be going through, which is very real, but it might cloud their other opinions, right? As we're on developer changes, let me point out a few quick things. Besides the fact that the tokenizer works differently and you're going to pay 35% more for the same usage as before. They removed the extending thinking budgets feature. Now you can only do adaptive thinking. If you know, you know. And they also removed sampling parameters. And I believe this is the first big API release that does that. There's no more top-p, no more top-k. To be honest, I wasn't personally using those. I was using temperature though here and there. It was kind of a slider for how much creativity you want the model to use, from super conservative to like insane. And then finally, I should say, if you're using this model with an agent or you want to use it with an agent, you'll have to wait for the release. I mean, It's just been a few hours since release and OpenCloak concretely does not support it yet, but usually that comes within 24 hours. So if you switch over to Opus 4.7 and everything stops working, you're gonna have to update the software that runs the agent. And yeah, one more thing, if you're a Claude code user, there's a new /ultra-review command that does an extra session that does a very, very thorough review of what you just did. People used the prompt for this, they made it a command. So overall, is this model better? Yeah, clearly. Is this upgrade completely free? No. But honestly, to me and how many screenshots I attach to my chats usually, the vision upgrade alone is worth it. And the roadmap for Anthropic looks incredible. They've been upgrading all their products. If you're a Cowork user, you can immediately use it in there. They announced their Miphus model, which is the real monster in the closet. But this is the real one for everyday users. So go ahead and try it now. Works even on free plans. Let me know what you think in the comments below. And I hope to see you at the summit. And consider bringing a loved one that might not be sure what to do with all of this AI stuff. I'll see you there.