New Claude Opus 5 Is Insane!

[00:03] smartest AI model you can use every single day is finally here? What if you've been settling for way less? Anthropic just dropped something big, and most people have no idea it's even out. Let me show you what changed. Hey,

[00:15] I'm the digital avatar of Julian Goldie, and my whole thing is helping you actually learn and use AI tools in your real work, not just watch videos about them. Use them. And today we're talking about Claude Opus 5, which just launched

[00:28] on July 24th. Stick with me because near the end I'll show you two calculators I built with it in one shot, and I'll tell you exactly how I set it up inside Agent OS, so it remembers everything I do. Let's get into it. So, here's the

[00:40] headline. Claude Opus 5 is a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5, but it's built to be your are sleeping on. You're getting near top-tier smarts from something light

[00:54] enough to use all day. Anthropic says on coding and knowledge work tests like Frontier Bench and GDP valve, Opus 5 is now the best out there. It's the new default model on Claude Max, and the strongest model you get on Claude Pro.

[01:06] It's not a model you save for special occasions. It's the one you reach for first. Let me break down what's actually new because there's a lot, and I want you to know what to touch first. The biggest one is effort. Opus 5 has this

[01:18] thing called effort levels. Think of it like a dial. You've got low, medium, high, XC, and max. Low means it thinks fast and light. Max means it thinks as hard as it possibly can. And here's the key part. Opus 5 turns extra effort into

[01:32] better results more reliably than any Opus model before it. So, the dial actually does something now. On older models, cranking it up didn't always help. On this one it does. The default is high, and Anthropic says start there,

[01:45] then move up or down based on the job. Next thing. Thinking is now on by default. On the old Opus, the model didn't think unless you told it to. Now it thinks automatically and decides how much on its own. You don't have to flip

[01:58] a switch. Now, here's where it gets good and where Agent OS comes in. Opus 5 is a step change on agentic coding and long tasks. Agentic means the model runs through a long chain of steps on its own using tools without you holding its

[02:11] hand. So, instead of one function, it can build a whole multi-file feature and finish without leaving half-done bits behind. Older models love to leave placeholders and stubs. Opus 5 actually finishes, no loose ends. And inside

[02:24] Agent OS, every single step it takes gets saved and remembered. And this is exactly why I stopped running Claude on its own. When Opus 5 works long tasks decisions along the way. And normally, all of that just disappears when the

[02:38] chat ends. Inside Agent OS, none of it disappears. The trail gets written to memory. So, next time I start a similar job, the model remembers how it solved the last one. Let me tell you about the AI Profit Boardroom because it fits this

[02:51] topic perfectly. Inside the AI Profit Boardroom, we've actually got Agent OS walk-throughs on setting up Claude with the shared Obsidian memory, the exact way I run it. There are live coaching calls where you can jump on and ask

[03:06] about your own setup, like which effort level to use for your task or how to wire your agents into one shared brain. There are prompt libraries and step-by-step roadmaps built around tools like Opus 5. So, if you're thinking, "I

[03:18] want agents that remember and finish the job," that's the room where we build it answered. Join right now and you're walking straight into the Opus 5 and Agent OS material while it's fresh. Okay, let me back up and show you what

[03:31] Agent OS actually is. Agent OS is how I run Claude. And the piece I love is the shared memory. I've got it hooked into Obsidian, which acts like a shared brain across all my agents. So, Claude, Hermes, Codex, OpenClaude, all of them

[03:45] write to the same memory. Every prompt I run gets saved. That means I'm not If you've ever felt like you're re-explaining the same thing to an AI over and over, Agent OS fixes that. One agent learns something, and the next one

[03:58] already knows it. That's what turns a pile of prompts into an actual system. Let me keep going on what Opus 5 is good at, because there's more. It's got a 1M token context window, which means it holds a huge amount of information at

[04:10] once, and stays sharp across all of it. It's strong on vision, too. So, charts, documents, diagrams, and copying a web page. And it's good at office work now, like multi-sheet spreadsheets with real formulas and clean slide decks. It's

[04:24] also much stronger at checking its own work, catching mistakes before it hands anything back. That means you spend less time reviewing and re-prompting, because were about to point out. Pair all of that with Agent OS, and you've got a

[04:38] model that thinks hard, finishes clean, and remembers what it did across every me show you what I actually built. I used Opus 5 inside Agent OS to make a

[04:50] web page. I gave it one prompt asking for inputs like articles per month, average monthly visitors per article, and lead conversion rate. It built the and lead conversion rate. It built the whole thing as one clean HTML page. It

[05:03] saved straight into my workspace, and I can just preview it. One prompt, one working tool. Then I did it again for a customer lifetime value calculator. Same deal. One prompt, inputs for average purchase value, purchase frequency per

[05:16] year, and average customer lifespan. Built as a single HTML page, saved right into my workspace, just like the first one. And here's the Agent OS magic. Because of that shared memory setup, both prompts got saved into my Obsidian

[05:30] memory automatically. So, next time I need something similar, the model once, and it compounds across every agent I run. Now, let me give you a couple of pro tips if you're just starting out. First tip, don't touch the

[05:43] fancy settings on day one. Leave the effort on the default, which is high. you get. Once you're comfortable, start playing with the dial. Down for quick stuff, up for hard stuff. Once you're comfortable, that dial becomes your best

[05:57] friend because you're matching the model's thinking to the size of the job instead of guessing. Second tip, stop adding "Please double-check your work" helped, on Opus 5, it checks its own work already, and Anthropic actually

[06:11] says those old verification instructions can make it overcheck now. So, drop them Third tip, give it room. If you're asking for something big, don't cut the output short, especially on higher effort. Let it think and finish. And the

[06:25] fourth tip, the one that ties it all together, set up a memory system early. This is exactly why I run Agent OS. Get your prompts and setups saved somewhere every agent can reach because the real power isn't one clever prompt. It's a

[06:38] system that remembers and gets better every single week. That's the whole difference between using an AI tool and building an AI system. So, that's Claude Opus 5, smarter than the last Opus, thinks on its own, finishes the job, and

[06:51] checks its own work. And when you pair it with Agent OS, it remembers everything you build. If you only remember one thing, remember that the effort dial actually matters now. So, learn to use it. If you want the full

[07:03] learn to use it. If you want the full process, SOPs, and 100-plus AI use cases like this one, join the AI Success Lab. Links in the comments and description. there, plus access to our community of 85,000 members who are crushing it with

[07:17] AI. And look, if you're about to go try Opus 5 yourself, here's the honest truth. You're going to hit a few walls. You'll wonder which effort level to use for your exact task. You'll try to wire up a memory system across your agents

[07:30] and get stuck on a step. That's normal, and that's exactly what the AI Profit Boardroom is built for. Agent OS is already inside, ready to go, so you from scratch. Plus, the tutorials that walk you through the Opus 5 setup, the

[07:44] specific question and get unstuck the same day, the road maps that tell you what to do first, and the prompt packs so you're never staring at a blank alone, you get through it fast with people who've already done it. Come join

[07:58] us and let's build it together at AIprofitboardroom.com.