Claude Opus 4.6 First Impressions & More AI News You Can Use

Welcome to another week in AI with serious improvements on all fronts, especially in the models that people seem to be using the most. Claude has a new flagship model with Opus 4.6. Spoiler alert, it's just better than anything I've seen yet. And OpenAI made their first move towards actually replacing human workers with an enterprise solution. And so much more in this week's episode of AI News You Can Use, the show that looks at all the releases, we filter for the ones that matter, and I get to refer them back to you. Let's begin. And we'll do that by talking about the big story this week. In my opinion, the thing that will touch most users that will have the most most effect on how people go about their everyday life. Claude Opus 4.6. There's been like an exodus of people moving away from ChatGPT recently, either to Gemini or to Claude. And the reason for that is that they've just been shipping features, new models, new application categories, and more. And this is another one of those cases. Claude was already considered best in class when it comes to building applications with their model Claude Opus 4.5, but now they upgraded it. Claude Opus 4.6. That's what we're looking at right here. In summary, the big change here is that this model is just so much more autonomous than anything before. Everything that has the word agentic in it, this is going to perform better. Agentic computer use, where it has its own machine and needs to figure things out. But even just agentic problem solving, where it kind of re-prompt itself in the process and tries to reach the goal that you give it in any given prompt. We actually had early access to this model. And honestly, from testing this thing and trying it out, my personal feeling was that this is going to be Opus 5. That's how much I liked how it behaved. And that's how much I thought of an improvement it was. Was, but it's actually just 4.6. Doesn't change anything about the fact that this is actually quite impressive and useful. Now, I just want to be clear, whenever we get early access to a model, there is no affiliation whatsoever. They're not sponsoring us. They're just saying, hey, we like what you make and we want to give you access to our products in advance so you can give a full opinion once they come out. That's it. In other words, if I don't like it, I can say it, then I will, as you might know by now. Okay, so enough talking. Let me actually show you an example of this performing versus the predecessor model. I'm just gonna run the same prompt in Opus 4.5 and Opus 4.6. It's a typical example we like to do here on the channel, especially in these news you can use roundups where it's about having a first look and getting a feeling for what's new out there. So we're gonna do a simple Pomodoro timer app with start, pause, and reset buttons and the ability to add multiple timers. This has actually always been a problem for all providers, especially when you ask for it in one prompt. I'm not saying models can do it. I'm saying they weren't consistent at doing it. Oftentimes a little bug snuck itself in because it had to build the Pomodoro timer, but then also it had to build an interface where multiple of them could be running at the same time. It's just a different level of complexity from just one timer. And when I ran this in Opus 4.5 right here in this chat, that's exactly what I encountered. This first timer works well, start, pause, reset, no problem. But when I want to add a new timer, let's say a break with 5 minutes, and I say add timer, this button just doesn't work. Now sure, I can fix it, but we're trying to one-shot something relatively simple to demo it in the video, which It just failed. Now, here's the result from 4.6. Not just that this is a bit more visually polished, clearly. I mean, just look at them side by side. This looks like a bit more of a children's application, but it looks good. This, I like it. That's just a subjective thing though. Let's look at the functionality. Start works well. Reset works well. Add timer. It even gave me some presets here. No problem. Look at that. I can add multiple. I can add a custom one. It just works. And although this is a simple example, it is representative of how these things are moving forward. So that's the Opus release. And they actually also released another model for the Claude Cowork application, which has seen minor updates over the past week. So I want to quickly talk about these. Besides the ability to work with this new Opus 4.6 model, they also added a thing that has been in Claude Code for a while now, but it really has been a power users only feature. It's something called plugins. And plugins are really a connection of multiple features we had already in here. If you're not familiar with all these, that's fine. But you might remember from previous episodes that Cloud offers these skills that are basically a pre-packaged workflow or guidance of some sort. So I showed you how to package brand guidelines in a skill. So every time you evoke that skill, it's gonna follow the exact brand guidelines that you taught all by just accessing that skill. Then there's also connectors where it connects to external applications. As I shared with you before, and I will share again, these are usually a bit wonky and unreliable, but they exist. And if you're very specific with what you're getting from the connector, It can work well. And then you can also do custom connectors through MCPs, which are basically just a standardized way to connect things to Claude. Now, all of that comes together in plugins. That's what plugins are. They just combine all of these different things and you kind of get this package deal of a skill that might really work well with a connector and you can just use it at the same time. From what I've heard, usually only developers have been using these. And if you're not a developer, it was kind of hard to find ways to use this. But now they brought it into here and While I can't make any concrete recommendations on what you could do, I would love to hear your opinion. If there's anybody who has been using these successfully, please leave a comment. I would imagine that something like the Skill with the brand guidelines paired with a connector into Canva or something like that could work well in tandem. Anyway, I just wanted to make this little segment to let you know that, yeah, they've been pushing the depth of Cowork with these plugins and with the new Opus model. And as I shared before, I think this product category is actually going to be the biggest one in the future because it's more about getting things done. Rather than helping you do things. So moving on to the next story, and this isn't fully out yet. I wish I could try this, but it's called OpenAI Frontier. And it's basically their best attempt at replacing enterprise workers. I mean, they're framing it up in a way that this is going to be helping enterprise workers. But really, it's connecting what enterprises already have with ChatGPT Enterprise. It's just a license that has more data security than the versions that most of us are using. But they pair them with custom agents and OpenAI agents that are basically set up to do certain functions over and over again. And then when they combine that with all of your business context and they give the agents the ability to really act within the company and they plug it into the internal systems of record, well, you're basically getting what an employee is supposed to be doing. This is the very first announcement in this direction that we see. And it's the very first consumer/enterprise level application from OpenAI that we really see pushing for this. And it's not even available yet to the wider public. It's just for enterprise customers and even within that only select ones. But I did want to highlight that here because this is sort of Claude Cowork, but from OpenAI, but more focused on enterprise. Again, I think this is going to be the main product category in AI. Hey, and by the way, if you enjoy content like this, don't forget to subscribe to the channel. It really helps us out. And now on to the next stories, which quick note, I recorded yesterday. So there's going to be a bit of a different look. I also went to the hairdresser. All right, let's move forward. So this next one is not necessarily something you can use right now, but it has been a huge story over the last week. So I want to briefly talk about it. Moldbook, the social network for AI agents, right? Where all these agents that we talked about on the channel last week went to discuss different topics like they would on a Reddit-style board. At this point, the story is already kind of old because people really stopped paying attention to this after it was exploited by different crypto people that were just shilling their coins on there. Long story short though, it's a vibe-coded Reddit-style application where people gave their AI bots, their agents, their— first it was called Claude bots. Now it's called Multibot. No, now it's OpenClaw. That's what it is right now. And they just discussed various things on here. So if you're watching this video, there's a high chance of you already having heard about this. But I just quickly wanted to give my opinion on this. And my opinion is that this actually doesn't differ much from AIs just talking to each other, which we've seen before. I mean, if you go into essentially every Twitter comment section, it's a bunch of AI replies and AI replies to other AI replies. And they're having discussions there. Now, this was way more inflammatory. They were talking about topics like taking over the world and how to handle their humans and forming religions, things like this. But at the end of the day, it was just ChatGPT talking to ChatGPT, or more accurately, Claude talking to Claude. No property of consciousness. That's what a lot of people ascribe to this. No property of, you know, them actually taking over. It's just a conversation between multiple chatbots. And a lot of that was even prompted by humans to start the conversation. I mean, people know how to get attention on the internet. And a lot of that is going to be by causing ruckus and, you know, talking about these inflammatory topics. And that's what happened here. It was interesting. It was fun. At this point in time, it kind of fizzled out, but I still wanted to talk about it because it was a really relevant story in AI. And also, it was a way to use these AI agents to communicate and went really viral. And I think it is important in the context of the other stories this week, which do touch on agentic models that go beyond what we've seen so far. So I wanted to include it. But now let's talk about the next piece of AI news you can use, because this one is a brand new category that is blossoming, and it's called Genie. Now, this one you might have heard about a few months ago when Google initially announced this Project Genie. This is their text-to-world model. If ChatGPT is text-to-text or text-to-website or text-to-table, this is text-to-world. Now, what that looks like in practice is something like these examples that you can see on screen right now. You really just start with a text prompt and it creates this interactive world that you can navigate. Like you would a computer game. Walk around in it, there's, you know, 3-dimensional properties to the whole thing. And it is really quite impressive. It is the most impressive thing we've seen out of this category so far. And that's why also this caught quite a bit of attention online, but then was quickly kind of overshadowed by all the Maltbook hype. But honestly, this is getting pretty darn good. Now, it's one of these technologies that is really like looking for a use case. It's impressive. I'm not exactly sure what it would be used for. I also brought up that question the last time I brought this up when they announced it and previous versions. But it's just getting really darn good. And I think we're going to see some surprising use cases for this in due time. Right now, this thing is only available for Ultra subscribers. That's their $200 plan. I guess on the trial is $125. And then fully, it's like $250. Anyway, if you're on the Ultra plan and are a US user with a US Google account, they're quite strict with that. That's why I couldn't get in. Then you can try this thing out. You can turn any famous painting or your favorite character into a little mini video game and run around or fly around or be a paper airplane, whatever it might be. And for the quick hits this week, let's start with Grok Imagine, the first big official image generator release from Grok. In summary, it's okay. I would say it's just another good AI image generator. We've seen so many of them now. I know for some of their marketing, they were doing more unhinged things. This is just a solid image and video generator. I suppose it's quite cheap. It's cheaper than the others. That's one thing. And it has super low latency. Those are two things that, you know, are an advantage. The demo videos look really solid. But once you test it yourself, and once you put it up against some of the competitors, like, na na banana, the ChatGPT model, Midjourney, as you can see here on the screen, it's okay. Okay, the next up, we have some updates from Google Gemini for Chrome. They're building it right into the browser. So they're really experimenting. I think that's the real story here. They're experimenting in every kind of way that you can imagine to integrate AI into existing products that people are using. You see inside of Google Sheets, you see inside of Google Search, Search Gmail now even in your Chrome browser. Not a game changer, just an ease of use update. Let's see if this sticks over time. This one is cool. Claude on Mars. And it kind of opens this debate of artificial intelligence actually being necessary to explore other planets because the latency is so high in communicating with those planets. This is not really something you can use. I just wanted to include it because I feel like things like this are very inspirational, at least to me. How far we've come as humanity and what's possible. If this inspires you too, I strongly recommend you check out the article. We'll put a link below. Next one are some lovable updates. I know a bunch of people go to this application to build sites. It's just the most aesthetically pleasing vibe coding application out there. It's also the lowest barrier to entry for beginners. So I mean, there's a reason why they're valued at $6.6 billion now. But they included a few things like planning mode, which is a core feature that we've seen inside of Cloud Code already. But it really helps if you're not sure what kind of app you want to build, you start with plan mode. And they also have automated testing now in there where you can just say simulate a user and test the website you built. It kind of does that. It works decently well. If you use Lovable, you'll love these updates. No pun intended. Next up, we have Codex, the application from OpenAI coming out. Honestly, this is sort of just like Claude Code existing in the terminal and then Claude Code being put inside of the web app. But it's sort of a weird release because people who are using Claude Code and comfortable with it, well, they usually don't mind using the terminal. And it's the same thing with this. Like if you're using application for developers that works in a terminal, you don't really need a much friendlier user interface. You need a different application that simplifies a bunch of other stuff too, like Claude has with their Cowork app. So yeah, this is just a case of OpenAI catching up and sort of having to do this. But I think it's a weird product category because I don't know, coding apps can just stay in the terminal, I think. And additionally to the Codex release, what OpenAI did is they released a new model that is only available within this app for now, GPT-5.3 Codex. This is a reaction to Opus 4.6. And while I haven't been able to get my hands on this yet, it's not available inside of my Codex, even when I activate the VPN. The reviews of this are very lukewarm up until now. I mean, obviously the benchmarks look good. There's some nice examples inside of the blog post. As per usual, you can see the difference between the last model over here to this model, you know, slightly better website design and other things. But with Claude, people are going crazy. People are like, Wow, I managed to write one prompt that one-shotted Pokémon. It just built Pokémon. It took it 90 minutes, but it just did it in one prompt. Cost 110K tokens, which in practice, I recalculated it, is roughly $3. But yeah, it just pushed what AI can do. And people are loving it. Whereas Codex is getting more of a neutral reaction. But that's me just parroting what people are saying online. I have to get my hands on this myself. I'll get back to you next week. But if you're a developer working with OpenAI models, you might want to check this out. And then lastly, Anthropic released a really very well-crafted guide on how to build skills. We talked about skills before, and basically you can package a certain behavior, a certain workflow, maybe brand guidelines, that's what I showed you in a previous episode of News You Can Use, into a skill so that it just remains consistent every time. But how to build them, how to really make them work well has been a bit of a trial and error type of situation. Well, now Anthropic released a full guide on it. So if you ever wondered about skills, there's a great guide for them now. If you haven't wondered, then this is probably not for you. But as per usual, you'll find the link in the description below. And that really rounds out this week. It's been colorful. There's a lot of releases and I hope you found something that is interesting or useful to you. And with that being said, my name is Igor Pogany and I hope you have a wonderful week.