Claude Sonnet 4.6: Opus-Level Performance at HALF the Price!

Enthropic dropped Sonic 4. 6 today and I've been going through the announcement and there's actually a lot here, more than a typical mid-tier model update. So, I want to walk through what actually matters, what the numbers mean, and whether you should care depending on how you use Claude yourself. Quick housekeeping, this is a Sonnet tier model, not Opus. It's the everyday workhorse tier, the one most people actually use. And the short version is they've closed a lot of the gap between Sonnet and Opus while keeping the price exactly the same. So, let's get into it. First thing worth knowing is that if you're on the free plan or even the pro one, Sonnet 4. 6 now is your default model as of today. You don't have to do anything. You already have it. The pricing on the API side is also unchanged. $3 per million input tokens and $15 per million output, same as 4. 5. So, if you're building on Claude, this is basically a free upgrade. Now, here's

the headline claim from Enthropic. Tasks that previously required Opus are now doable with Sonic. And that's a big deal because Opus is substantially more expensive. If that claim holds up in practice and based on the customer quotes in the announcement, a lot of companies are saying it does, that changes the economics of building with cloud significantly. The other headline is a 1 million token context window currently in beta. To put that in concrete terms, that's enough to hold an entire codebase, a stack of lengthy legal contracts, or dozens of research papers in a single request. Coding is where Enthropic is leaning hardest with this release, and the numbers they're sharing are pretty interesting. In internal testing on Cloud Code, that's their command line coding tool, developers preferred Sonnet 4. 6 over Sonnet 4. 5 about 70% of the time. That's a wide range for a preference test. But the most important part is this. Users preferred Sonic 4. 6 over Opus 4. 5, their Frontier model, which was dropped just a few months ago, 59% of the time. Why? The feedback is pretty consistent across different users. It's less prone to overengineering. It actually reads the context before it starts changing things, which sounds obvious, but was apparently a real frustration with earlier models. It consolidates shared logic instead of duplicating it everywhere. And there are fewer false claims of success, which means more consistent follow-through on multi-step tasks. The last one is huge. If you ever use a coding agent for anything longer than a oneshot task, the failure mode of it telling you that it's done and it isn't is one of the most annoying things about working with these models. On the software engineering bench verified, which is the main benchmark for real world coding tasks, this new model comes in at 79. 6, 6, which is a little bit below the Opus 4. 5, which sits at 80. 9%, but it's an improvement from the Sonnet 4. 5 version, which was currently sitting at 77. 2%. Before we continue, if you're watching these AI releases every week and wondering how to actually use them in your work, this is for you. Outskill is the first ever AI focused educational platform to accelerate AI learning for people like you and me. They're hosting the 2-day AI mastermind training live this Saturday and Sunday between 10:00 a. m. to 7:00 p. m. Eastern Standard Time on both days. And right now is the perfect moment to join because you can get in absolutely free for the next 48 hours. This 16 hours AI training has already built 10 million plus AI first professionals worldwide and is rated 4. 9 out of five on Trustpilot. People from marketing, finance, operations, product and engineering join this just because this is something which is not specific to any industry but is now needed across every one of them. This is where you learn how to build AI agents that plan, write, execute and report for you. Automate workflows that run even while you sleep. Connect tools like sheets, notion, CRM, and email to create profitable system. and learn how to use AI to save hours every week and get an unfair advantage at work. Not just that, you'll also learn how to profit from these skills. People from this very training have launched AI powered services that bring in 3,000 to $4,000 weekly just by applying the systems they're taught. To kickstart your year, you also get premium lifetime bonuses like the AI prompt bible, the AI profit road map, and your personalized AI toolkit builder. And here's the most interesting part. When you sign up, you'll get access to the 2026 AI survival hackbook, a comprehensive compilation of the upcoming AI shift in 2026 and the practical steps you can take to be prepared. Seats are limited. Use the link in the description to join. Also, join the WhatsApp community to get updates before the event. All right, let's get back into the video to see how

powerful this new model is, Cloud Sonic 4. 6. I'm going to test it against 4. 5 because apparently people are preferring to switch over from cloud opus 4. 5 to Sonic 4. 6 and remember this is a much cheaper model compared to Opus 4. 5. But if this new model is not only able to produce similar result but may maybe hopefully a little bit better results then you know that Claude Sonic 4. 6 is going to become a new model that most people will switch over to. So we can see I've asked these models to create a bowling game with all these features. I'm not sure if all of these features will be included, especially since I'm only doing a one single prompt and trying to one prompt code this. So, we can compare the results. So, this version over here is the model from Cloud Sonnet 4. 6. Visually speaking, this looks pretty good. Like, I can aim and everything like that. I can curve the ball a little, I guess, if I throw it over there. And then this is the visual of the Cloud Opus 4. 5. From the get- go, I'm going to say 4. 6 six wins over here because this kind of looks like footballs over there. And then, you know, it doesn't have that hoverable feature. Like I don't even know if I can move this around, but yeah, actually I could. I have this aim position over here. That will kind of change it versus Cloud Opus 4. 6. Clearly, it's using the mouse features. Let's test it out. So, I'm going to throw one here. And then I have this meter, the power meter. Look at the animation, guys, over here. This is so good compared to like what we've seen previous models. So we can see that. Okay. Split for knockdown. It was a little bit animation on that. Let's try this. I don't know what this is going to do. And then I'll put it like a little bit less speed. Miss. Okay. Five. Okay. Interesting. Obviously, these are not perfect. And remember, this is one shot. Let's compare this to the Opus 4. 5 model. So, I can change my aim. Let me make this full screen. Change my aim a little. Also, the visual thing like why is I have to scroll down to make it fit. But let's change it. Hook right. Stuff like that. Power. Power meter. Okay. What the hell? These models are kind of crazy. But the power meter on this one compared to the Opus 4. 6. You can clearly tell that this new model is much better at coding compared to previous versions. And yeah, I can see why developers are trying to use this new model. They get access to good coding capabilities at a cheaper cost. So it makes sense to do that, right? So I understand why the people are switching over to this. But we also wanted to compare against the

new Claude Opus 4. 6. So this is a new model that they dropped couple of weeks ago. Sorry, not even a couple of weeks, maybe a couple of days ago, maybe last week. What I've asked it to do, create a international space station tracker. So both of these visuals that we're seeing are going to be from both of these new models. So sonnet 4. 6 versus cloud opus 4. 6. Obviously we know that this cloud opus 4. 6 is a workhorse model meaning this is supposed to be the one that people would do uh that people would use for everyday coding, everyday software engineering tasks, more higherend coding things. And the sonet 4. 6 six is kind of a I would say is a generalist model that's supposed to help you not only do everyday tasks but now apparently also help you code. So we'll compare these results as well. So we're going to look at the results of the ISS live tracker from the Sonnet 4. 6. So this is not bad. Like we have our tracker over here. Our map kind of looks a little bit off everything like that. Orbital velocity. I'm not 100% sure if these metrics are accurate but I think it might be. Latitude longitude. can track it. It's changing altitude, velocity, daylight, everything like that. Current crew, seven members. So, we have people from NASA on there as well from the Rosscosmus. Okay. And I can view all Oh, this is kind of cool. I can go into the crew section. This is really cool, guys. Like, it's adding all these features. And I just give a simple prompt, right? I haven't given it any more details. This is a feature that I added by itself. station specs um mass length pressurized volume first launch. This is kind of cool. Okay, past predictions today visible past predictions. So, it also had a prediction model of when we're able to see it, everything like that. Past prediction, viewing tips. So, this is really sick. Obviously, this is not as beautiful compared to the Opus 4. 6, but remember guys, we're not trying to get comparable results. We're trying to get something that's similar, not 100% accurate cuz this model is definitely more stronger when it comes to coding, but the results are quite similar. Like we have this, the visual layout of this is definitely much better compared to the other one. We have our crew abroad here. Looks like they are pretty accurate compared to each. We have everything like that. This one also added days, how long they've been on expedition for, latitude, longitude. So altitude over here is 408. Over here we are seeing 406. So, not too bad. And then we also have [snorts] in this the cloud opus 4. 6. We also have station specifications at the bottom over here. Uh, but you know, I personally I like this. You know, I'm able to go through different dashboards here. It shows that the model did some thinking and organize it a little bit better. Obviously, the visuals are not as strong as compared to the one over here, but I'm still going to give both of these models a great job. Um, but yeah, I can see why people are switching over to Cloud Sonnet 4. 6. And you know, this is comparing it against the 4. 6 model, which is probably one of the most frontier models at the moment. So, good job done by the

Enthropic team. Make sure to subscribe to our channel. We do real tests, not just headlines. Make sure you're also subscribed to the world of AI. And don't forget to check out our newsletter for deeper breakdowns you won't see on YouTube. And I'm growing my Twitter following, so make sure you follow me on Twitter as well. Hope you guys enjoyed today's video and I'll see you in the next