Claude Sonnet 4.5 - Why You Need To Use It

0:00Cloud Sonnet just released and here's what you need to know. This has been amazing and it's the moment we've all been waiting for. Maybe you've been using Cloud Code for the last few weeks and you've noticed that it's not quite the same as it used to be. Well, today Anthropic just released Sonnet 4.5, and supposedly it's supposed to be one of the best agentic coding models compared to Claude, uh, compared to, um, GPT5, compared to Codeex, compared to Grock, uh, GLM, and uh, other Agentic coding agents that are out there. And I've been using Claude to build out our side projects over the last 65 70 days, showing um people on this channel that hey, if you have no experience, you can

0:46build something substantial, too. So, just going over it, what we're building and what we're working on is um pray about it is a prayer app turning prayer into actions. And the most recent project that I haven't really talked about with you guys is um Perfecto Pickic, which is built for real estate agents using the Nano Banana as the backend Gemini 2.5 Flash to restore um photography photos for your listings or if you're working in a wholesale and you um you know want to show improvements whether it's virtual staging and whatever. Um that's something that we just launched today uh with a business partner. So, I can talk more about that later. And then our first app that we ever built on the channel was Easy Flip.

1:31This allows people to take pictures of items and automatically list them onto marketplaces like eBay, Poshmark, and Merkari. So, those are the three things we've worked on, all built with Claude Code. And just wanted to show you what this is going to impact how this is going to impact my workflow and hopefully can impact yours because the thing that I've been struggling with recently and if I was going back in time and and going through this, you know, my first few weeks, I would have never known. But now that I've been working with clawed code for four to six to eight hours every single day over the last uh two months, um I now notice when it hallucinates, when there's a degradation in the output, when it just goes off on tangents writing code and forgets about

2:18the subject that we were talking about and I'm like, what is going on? Maybe you've seen that before as well. But now today, my friends, I've been using it a little bit today, working on a side project, and I've just noticed it's faster. Like straight up, it's just faster. Um, it's able to like almost work on multiple web pages at once, which I've never even seen ever before. It's always worked on like the landing page and now it's working on the dashboard. It's like it's able to now work on the landing page and the dashboard. I don't know how, but it's just faster. So let's walk through um you know supposedly supposed to be the best coding model in the world right now and this is going to be default for people that are using um claude um I use

3:04claude in cursor so um I was able to just update it um and you can see here if you just go forward/model you can see that claude sonnet 4.5 is the default model so pretty cool um like I said I've only used it for you know, 30 minutes or so. But I've just noticed like, hey, it's faster. It's snappier. Um, doesn't mean it's amazing just because of that. But, um, yeah, so pretty cool. Um, I won't read through all this text here, but I think that these visual graphs can really help paint a better picture. So, software engineering bench, um, 7 77.2% versus the Opus 4.1 model. Notoriously, the Opus 4.1 model was one of the better

3:49models to code with. um over the last few months and that's what I had been using and building on. Um but I've just noticed a degradation in quality from time to time. Um and that's just coming from me with two months of experience. So take that for what you will, but if you see other people that are also making content in this space, um they notice the same thing as well. So you got to stay on track. You got to keep on track. Make sure it's not just writing code everywhere. Um so Sonet 4 represents a significant leap forward. um a benchmark that tested AM models in real world computer tasks. Sonnet 4.5 now leads at 61.4%. Just 4 months ago, Sonnet 4 held the lead at 42.2%. Um Cloud Chrome extension puts these upgraded capabilities to use. So you

4:36could watch the demo here. I'll link this down below as well. The model also shows improved capabilities on a broad range of evaluations including reasoning and math. And this is a really good one. really detailed. Um, so Agentic Coding 77.2% versus Opus 4.1 74.5% and then with parallel test time compute 82% versus Claude Opus 79.4%. And if you're going to be talking about Gemini or GPT5, um, not substantially better, but there is an increase in performance. Agentic terminal coding 50% versus GPT5 at 43.8. 8% even Claude Sonnet 36.4%. So a nice leap forward. Um Gemini uh 2.5

5:24Pro kind of falling behind there. Um Agentic tool use um telecom 98% airline 70% retail 86.2%. So a nice leap forward compared to Opus. Computer computer use 61.4% versus 44.4% on Opus. Uh, high school math competition, 100% on Python, 87% no tools versus the 78% on Opus. Graduate level reasoning, 83.4%, 81% versus Opus. These things are getting so much smarter now. And today is the day to start that project you've been wanting to start, guys, cuz these things are getting better. They're getting faster. Like I said, going back 65 days ago, not really knowing what I was

6:10doing, but I started messy. I figured out what made a good app, what I needed to build around um the idea that I wanted to build the app, and now this just helps you put that together quicker, faster, easier. Um now, better time than now to build the app that you're wanting to build. Comment down below if you have questions. I'd love to help you guys get started. Maybe you've been looking from the outside in. You're seeing all these cool things happen. Doesn't mean that you can't start. Because I was the same way for about a year and a half. Once GPT3 came out, um Claude Sonnet 3.5 came out. I was just watching people build cool stuff and I was like, "That's amazing, but it's not for me because I don't know how to do it." I I abolished that minimal thinking

6:57that uh you know poor train of thought and uh um closed mindset and just started messy and started dirty and just built. That's where Easy Flip came in. I wanted to document that journey. It's done. It's finished. We're making a $1,000 MR. While it's not a ton, it still shows and validates my idea as people want it and that's great. Um, and now we're going to be documenting more projects on the channel, and I can't wait to share with you all because I know that you can do it. The world needs what you're trying to build. So, build it. Use Sonnet 4.5. Um, and get going. Get going. So, um, comment down below, what are you working on? Or maybe comment down below why haven't you started working on your project. So

7:42experts in finance, law, medicine, and STEM found sonnet 4.5 shows dramatically better domain specific knowledge and reasoning compared to older models including opus 4.1. Finance 16k thinking don't know what that means but 72% versus 68%. Okay, so opus was 60% versus the 72%. And then law uh sonnet 4.5 65% versus opus medicine 61% versus opus 53% and stem. So gradual leaps forward. That's awesome. Um let's see where was it? Okay this was the main one that I wanted to talk about was misaligned behavior scores. So this the

8:29the lower the score the better. So, misaligned behavior scores from automated behavioral auditor. Lower is better. Misaligned behaviors include but are not limited to deception, um power seeking, encouragement of delusions in compliance with harmful system prompts. So, basically, hallucinations. Um you can see Sonic 4.5 um scored really well on this. Scored really well. Um GBT5 coming in at a close second. Um, but yeah, that's really cool because that's the thing that had been something that I had always had to check just to make sure Claude was doing what I wanted to do. Um, you know, maybe you've had um a a contact a context window get compacted

9:15and it gets like cut off in the middle of our our tasks that we're trying to complete and then it starts doing stuff that are just like it's like what are you doing? It's it's like working on the landing page. It's like no, I didn't want that. we were in the middle of working on this markdown file. Why are you doing this all of a sudden? So, I'm hoping and in and believing that we're going to see less and less of that um and less delusions, less craziness from Claude. Um and that kind of chart proves that in the testing. So, let's see the Claude SDK agent. So similar um to like OpenAI SDK agents um it is the same infrastructure that powers cloud code the agent SDK but it shows

10:02impressive benefits for a very wide variety of tasks not just coding so you could build your own agents today we built cloud code because the tool we wanted didn't exist yet the agent SDK gives you the same foundation to build something just as capable for whatever problem you're solving. So very cool and um I'm loving it. I love these updates. I think it's a great time to be alive to be building the project that you want to build. I mean I knew nothing 65 days ago and we've built perfect pick. We've built Easy Flip and the next project is going to be that Pray About It app. Um so a great time to be alive y'all. a great time to be alive building what you want to build. And I just wanted to give you that update because I think it is a substantial update for people that want

10:48to get building on their idea, but maybe they're limited by, you know, I don't know how to use these tools and I don't know where to start. Well, if you just look back, I didn't know either. I started messy. I just made mistakes. Bugs were popping up. But you kind of have to have that addiction to overcoming problems and problem solving. gives you like a nice dopamine high or a nice push in the right direction and a sense of fulfillment that you're overcoming these problems. You never knew how to fix them, but you're making it work and you're making it happen. And that's why I'm going to keep showing you on the channel and just building in public and showing that you can do it as well. Uh comment down below what are you building? What are you working on? Um I would love to know you guys. We'll see you in the next video. Bye-bye.