Everyday AI Podcast – An AI and ChatGPT Podcast - Ep 816: ChatGPT Work and GPT-5.6 Sol: What’s New, 5 Overlooked Features and 1 Hot Take

Starting point is 00:00:00 This is the Everyday AI Show, the Everyday Podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Most AI tools at work don't know anything about you. Every time you're stuck re-explaining your projects, your team, what actually matters. Slack just fix that. The all-new Slackbot is a personal AI agent built into Slack, and it starts with your context, your messages, files and channels, translation, an AI that already knows how you work. Check it out at Slack.com.

Starting point is 00:00:47 Yeah, yeah, yeah. OpenAI's new GPT56 Seoul is amazing and the best and most practical AI model in the world right now. But they actually just pulled off one of the smartest naming tricks I've seen in a while. See, they took the popular agenic coding platform codex, soften the edges, got rid of that scary word code, and just repackaged it as chat GPT work, the super app. But everyone is rightfully paying attention to the new model itself, GPD 5.6, Sol, OpenAI's flagship release built for longer, cheaper, and more reliable execution of all type of work outputs. This sounds like one launch of two related products,

Starting point is 00:01:36 But it's really a model story, a platform story, and a competitive story all wrapped up in one. And there's some details that I think people are overlooking. So that's exactly what we're going to be tackling today on everyday AI. So here is the big picture. Open AI just released their new 5.6 models. Yeah, GPD 5.6 models. There's multiple. And a product they're calling chat GPT work or the.

Starting point is 00:02:06 the super app. So the models themselves, well, by benchmarks and my limited to you so far, look and feel like real upgrades built for long running, hands-off professional work. So chat GPT work because you're going to be hearing me say work and the words chat GPT work a lot. It's, well, it's just kind of repackaging of some of the tools you didn't know you had inside of chat GPT. And in case you did miss it, it's been a very busy like 36 hours for Open AI because not only did they come out with probably now the best model in the world, the best harness and making it more accessible for more people, but they also brought what I think could be more impactful in the long run yesterday. No,

Starting point is 00:02:55 two days ago. My gosh, time's flying by. The new GPT Live model, which brings that duplex two-way voice interaction. But the big thing is GPD 5, 6. So on today's show, you're going to learn why chat GPT work on the web is mostly just pretty packaging and where the real upgrade actually lives. I'm going to tell you five overlooked features that people aren't really focusing on and they probably should, including the full merge super app and the quiet death of Atlas. But don't worry, it's a good thing. And I'm going to give you my hot test. Open AI's positioning play and why I think Anthropic should be very nervous of a model that's not GPD 5,6, soul. Yeah, a different opening eye model is going to cause anthropic headaches.

Starting point is 00:03:49 Let's get into it. Welcome to Everyday AI. If you're new here, my name's Jordan Wilson and, well, do this every day. It's for you. It's your daily live stream podcast and free daily newsletter, unedited, unscripted, helping business leaders like you and me keep up with the AI updates. I go through sleep deprivation so you don't have to. You get to benefit from just the good stuff to grow your company and your career. So it starts here, but make sure you go to your everyday AI.com. Sign up for the free daily newsletter. We're going to be recapping all of my

Starting point is 00:04:19 random ramblings into the too long didn't listen version. So make sure you go do that and check out everything else that's happening in the world of AI in our newsletter. All right. So let's get to it. We're talking about chat, GBT work. The new GBT 56, soul. I'm going to tell you what's new, five overlipped features and one hot take. So GPT 56 Seoul. If you miss the boat or maybe you've been on a long vacation and haven't listened to the podcast in two weeks, opening I already announced this. This wasn't new. We knew it was coming. They actually announced it about two weeks ago, but they had to work with the federal government now that we have kind of the permission slip, you know, way of doing AI here in the U.S.

Starting point is 00:05:03 because of Anthropic. But here it is. Brand new model. But the difference is, well, it is actually maybe a little more confusing in the long run. Because before we just had things like GPT-5-5, right? And then you had different levels of reasoning and all that. So now we still have the different levels of reasoning. We still have the number, the 5-6, but now we also have Seoul, which is the big model.

Starting point is 00:05:32 we have Terra, which is the medium model, and then we have Luna, right? So sun, earth, moon, which is great. And I think that that's going to carry on. So let's talk a little bit about the new model. And this is from Open AI's release. So they said we're launching the GPD 56 family of models for general availability following our limited preview. Our new flagship sole alongside Tara, a balanced model for everyday work and Luna, our most cost-efficient model. GPD 56 Sol sets a new standard for both intelligence and efficiency, achieving state-of-the-art results across coding, knowledge work, cybersecurity, and science, while outperforming previous and competing frontier models with fewer tokens and at a lower estimated cost. The result is stronger performance per dollar, more successful work for the same spend,

Starting point is 00:06:20 or comparable results at a lower total cost. We also introduce a new way to accelerate the most demanding work. Ultra is our highest capability setting. coordinating multiple agents across parallel work streams to finish complex tasks faster. Stronger computer use and design judgment make GPD 56 sole our most polished collaborator yet, helping it inspect, refine, and deliver ready to use results. All right. So we're going to talk a little bit more about GPD 56.

Starting point is 00:06:51 So I'm going to do a little bit comparison, not too much on this show, but obviously Fable 5, right? We have to talk about those two models together when we go over the bench. here in a couple of minutes, but I want to quickly tell you a little bit about the other big release. So we knew GPT 5-6 Seoul was coming out yesterday on Thursday. What we didn't necessarily know is, well, we were essentially getting the super app, right? Internally, they've kind of saying it's the merge, but this is the super app. So this is kind of the combination of chat GPT codex, which has been on fire the last five months and the Atlas browser. So the Atlas browser is no more.

Starting point is 00:07:34 Now, Codex is chat GPT and chat GPT is codex. It's actually both an amazing play, just simply calling it chat GPT work, but also kind of confusing with how it rolled out because if you go download codex or if you go update codex, you know, now it says, you know, chat GPT, but you can change it. You know, you can say, hey, I want it to be the codex icon and still call it codex, but it's the same thing. Right. But kind of how Open AI says put their post out about chat work.

Starting point is 00:08:13 They said, powered by GVT56, chat GVT work brings together context from your team's tools to turn scattered notes, drafts, and ideas into finished work. And it keeps projects moving while you stay in control. They say, Chad ChaptiPT work gathers context, plans the approach, and takes action across your tools, files, and desktop apps to create published spreadsheets, docs, and slides. Chad Chapti can turn context from your tools and files into polished documents, presentations, and analysis to better, that better follow your templates and preferred formats. And then they say sites, yeah, updates on sites. And I'm going to share about that more in a minute. Sites let you turn ideas, plans, and data into interactive websites and web apps. that are easy to share with your team.

Starting point is 00:08:58 Create dashboards, project trackers, launch calendars, prototypes, reports, and more, and keep them up to date as information changes. So then talking about the unified plugins, they say with more than 1,400 plugins available, ChadGPD can pull context from the tools and workflows you already use to help move projects forward. They say create one time recurring tasks, monitor updates, and check progress from your phone while you're away from your doubt. and I'm going to be talking about those chat chbt remote updates because I think they're really good. Then the new side by side in chat chbtee, which I like, it's the new built-in browser experience

Starting point is 00:09:39 in the chat chvety desktop app makes it easier to work across your tools, files, and accounts with support for multiple tabs and richer agentic workflows. And I'm going to share at the end some of my tips on the best way to do that because that's, For me, the biggest unlock. A very small piece of this is the biggest unlock. And then review the approach before work begins. In plan mode, chat TVT gathers context, ask questions, and creates a step-by-step plan. You can suggest changes or approve the plan to start the work.

Starting point is 00:10:12 All right. So let me translate all of that marketing speak. It's just codex. That's it. There's really nothing new. So I don't know if most people remember that. But Open AI added a feature about two weeks ago, or sorry, about two months ago. Essentially, it allowed you to choose how you wanted codex to display.

Starting point is 00:10:36 So you could either have a more technical display or something they just called for everyday work. So you could just choose how much technical detail codex showed when it was working. So did you want to see all the tokens spinning and all the Python's calling and all the tools spinning off the shelf, right? If so, you would choose the coding approach, which is technically now codex. If you didn't, if you just wanted to more minimalistic and you didn't really care about what was going on under the hood, well, that's now just called chat GPT work. I spent so much time and I'm like, this can't be true, both on the web, because they did the same thing on the web. On the web, now you have chat GPT and there's a toggle for chat and a toggle for

Starting point is 00:11:21 work. And, you know, on the web, at least, chat GPT work looks a little different. There's technically no real new functionality aside from being powered by the new model. The plugins are unified, but there's no new anything, right? So you look at it and you're like, oh my gosh, this is going to be amazing. What's new here? All this is, which is smart. Don't get me wrong, right? I just was like, okay, I just, you know, read the blog post six times, watch everything. every single video that Open AI put out, you know, played with every feature. And I'm like, oh, it's literally the same. But I don't think that's a bad thing.

Starting point is 00:12:00 I think it's smart what opening I did here. Because if I'm being honest, even power users of Chad GPT, I could go in. I feel show them things. And then people would be like, wait, what? Like, where'd you find that setting? Right. There's literally be some things that were some of the best parts of Chad Chbbtee that you couldn't even find, right?

Starting point is 00:12:19 You had to go into like a setting. like four or five rows in and it's like oh my gosh you can do that in chat gvt it's like yeah you can't um you know like even like scheduled tasks they used to uh they took that away for a couple of months but you could still access it via a URL or by clicking like four or five links deep inside of a setting panel so you're gonna probably be hearing a lot of chatter about the new chat jvt work i do think it's smart so on the web it's just some more uh you know it does look like more codex on the web but you don't really have the codex features per se. But on the desktop,

Starting point is 00:12:59 it's just codex. Chat Chb-T work is codex and you can choose to call it chat Chb-T or call it codex, but you will have a kind of a switcher at the top and you can switch between chat chvety work and chat chav-tic-v-tac-tcodex, but nothing really changes. But like I said, I think it's smart. And here's the play. It's still baffled me.

Starting point is 00:13:28 How many smart people I've talked to over the last five or six months about Codex, right? Because on one hand, yes, it is an extremely powerful, agentic platform that you can automate your life. You can code. You can build things. Right. And that's intimidating for people. Right.

Starting point is 00:13:44 And I've been saying this on the show. You know, I've been, you know, Captain Codex since February. And it absolutely crushes every. competitor and it's not even close. It's way better than clawed desktop, Claude code, Claude, code work. It's better than cursor. I think cursor's probably in second place. It's better than anti-gravity, Gemini. It is the best desktop AI app and it is not even close. But the problem is is the word, codex, right? People said, oh, I'm not a coder. I've had so many recent conversations. People who are even somewhat technical, but they're not coders.

Starting point is 00:14:20 And they're like, oh, I can't use codex. And I'm like, you absolutely have to. So from that aspect. It's a very smart play. They literally just slapped the word work on, didn't really change anything, just shows fewer details. That's it. All right. So don't get confused when you're like, what's this new Chad TPT work? It's really nothing new. All right. So yeah, here on my screen. I have the Chad Chbitty work on the web just looks kind of like Codex light, but it doesn't really do anything new or different that you couldn't previously do on the web inside of Chad Chivity. So yeah, the web version just bundles agent work, deep research projects, connected apps.

Starting point is 00:14:58 So it just brought some of the more technically advanced features, the ones that I used all the time. And it kind of gave it its own dedicated panel in the work. You can still do all those things in the normal chat panel. It just, you know, makes the chat panel maybe a little bit easier and less intimidating with, you know, fewer interfaces. And then the work panel just makes it easier. And then it suggests different things. All right. And then, well, there's one new thing.

Starting point is 00:15:27 There's a nice little toggle, right? Inside the chat, chat, GPT desktop app. Yeah, there's nothing really new there. But Atlas is built in. Sorry, I take that back. Atlas is now built into the browser. And it is really good. It is improved.

Starting point is 00:15:44 And I'm going to give you here in a couple of minutes. When I go over my five overlooked features, one of them is going to hopefully, I think, change the way a lot of people work. All right, so more on GPD 5, 6, you know, got to get into the benchmarks, the comparisons, all that. But I need a quick water break. And we have to quickly hear a word from our partners. Here's what most AI tools still can do.

Starting point is 00:16:11 Work outside their own little box. The all-new Slackbot just change that. It's your AI teammate inside Slack. And now it can read, write, and act across the other apps your team already uses. No more tool switching. No more re-explaining yourself every time you open a new tab. One ops team at Engine says the summary feature alone saves them 15 to 20 minutes a use. See what Slackbot can do at Slack.com.

Starting point is 00:16:39 All right, so a little bit more on the new models. So, GPD 5, 6, here's the basics. There's three models, Sol, Terra, Luna. Sol is the flagship. It is aimed at the heavy users, coding, knowledge, work. cybersecurity and science. If you are someone that's using AI for three, five, six, 12 hours a day, you're probably going to want to use Seoul because you will be able to take advantage of

Starting point is 00:17:07 those full capabilities. We always talk about the capability gap, right? If you're doing lighter work, maybe you can use Terra, right? That is the more everyday work at a lower cost. And then Luna is the fastest, cheapest tier. So, yeah, that's our new three naming tier. And then on top of that, you have Soul Ultra as well. So let's look at that.

Starting point is 00:17:33 So with Soul, so we're going to get into the benchmarks here. It did extremely well. So even on a newer benchmark, which I really like, if you've heard of Humanity's last exam, it's essentially all these questions that are not in training data. So there's a similar one called agents last. exam. And it's been one of the more popular in trending benchmarks of 2026. And it just crushed the competition. So, you know, and even if you're sitting there and you're like, I don't need agents. I'm not a coder. Yes, you do and yes, you will. You just don't know it yet. And I'm being

Starting point is 00:18:12 dead serious, right? The future of work. If you are a knowledge worker sitting in front of the computer, you are an agent orchestrator. It's how I've been. doing my work since February since Codex came out. You know, constantly providing a lot of feedback on the front end, you know, keeping an eye on your agents, seeing what they return you. That's your work. You know, you polish it a little bit, but you're able to get four, five, six, 10 times done that of the type of work that you would normally do and, you know,

Starting point is 00:18:45 unlocking a whole new skill set that you never thought you could do. So, but the company's core pitch here is just performance per dollar. going to get into that in some of the charts. And then there is technically a tier-ish on top of Sol, and that's Sol Ultra. And that's just when they run essentially four kind of sub-agents in parallel, all of the highest setting, and you kind of get the best of the best. So Ultra does kind of coordinate those four agents in parallel by default to finish those demanding tasks faster. We've gone over pricing before, but Seoul is $5 input and $30, per outputs for million tokens.

Starting point is 00:19:26 So that's just if you're paying on the API side, right? So if you're in, if you have a Chad GPT subscription, FYI, that now means you have a Chad GPT work or a Codex subscription. It's the same thing. It shares those limits. So obviously you're not, you know, having to worry about that. But the pricing is extremely important. And then obviously the cheaper per task can still drain plans quickly.

Starting point is 00:19:51 So, you know, if you are using, the ultra, sole ultra, that's going to eat, eat things away. All right. Let's talk a little bit about some benchmarks. We're not going to go too deep into it, but it's important to talk about. So I think probably the best single source for benchmarks is artificial analysis. So it is conglomerate. It just brings in, you know, different benchmarks all at once.

Starting point is 00:20:21 So it averages all these different benchmarks. So it's one of the best sources to look at. So GPD 56 sole max, pretty much on par with Claude Fable 5. It's a one point difference, which is pretty minuscule, right? I'd say you can start power users can start noticing at about a three point difference. Average users, maybe like a six point difference, and they're not going to tell the difference. But, you know, if you are a heavy user, you know, if you are a heavy difference, user, you're not going to be able to tell the difference between a 59 and a 60 on artificial

Starting point is 00:20:58 analysis intelligence index. That means they are essentially one A and one A.A.B. It's not one A and one A B. It's one A.A.B. But that's just on that, the artificial analysis intelligence benchmark. Here's where it gets interesting on, they have, essentially they have three main benchmarks. I usually just talk about the intelligence index because that brings a little bit of everything. But in this case, when we're talking about work, knowledge work, becoming more and more agentic. I think that's actually when the other two artificial analysis benchmarks become more important. And that is their artificial analysis coding index, which GPD 56 soul beats Fable 5. All right.

Starting point is 00:21:40 So it beats it on coding. And on the agentic index, 55 also beats it. So essentially, I would argue by most standards, that. would mean that GPD 5-6 sole is better. I mean, we'll see when all the benchmarks come out in the LM arena. So what do humans prefer? Right. And, you know, Open AI has said they're working on front end design.

Starting point is 00:22:06 I know so much, which I never understood, right? Because so much of like vibes that are out there for models, it's always like, okay, I created this video game and like, look at how good it looks. It's like, okay, well, maybe like what, 0.001%? of the world uses models for that, right? I use models for knowledge work. I use models for personalizing research, carrying my context over from different apps,

Starting point is 00:22:31 you know, building decks, building little sites that I can share, right? You know, helping me grow my business, right? I'm not creating three Gs worlds. You know, I do a little bit of design things, maybe for some of the apps that I do build. But you know, I think it's important to look these three benchmarks in tandem. But the one that is the most telling. And this is,

Starting point is 00:22:57 we're going to get to this in a little bit more is the price. So for artificial analysis, this is essentially the cost per intelligence, right? Because these models can all get the tasks done. But well, how much does it cost? And right now, GPD 5,6, I will argue, is a slightly better model. If you look at all these, you know, the three different artificial analysis indexes that are the most important, you know, they're one point behind on one winning the other two, but they're doing it at, yeah, a dollar per task, a dollar in four cents versus Fable five, $2.75. Yeah. Anthropic. All right. So we went over Chad GBT work. We went over Codex. Now I'm going to give you the five overlooked features. and my one hot take. All right. Overlook feature number one, even though Open AI didn't say it out loud,

Starting point is 00:23:55 this is the super app, right? I think there are a lot of super app, you know, rah, back in like March, April, May. And at the time, Open AI said, well, Codex is the super app, right?

Starting point is 00:24:12 They said, you know, they said, we're building it in disguise. But this is it. But it's bringing together, chat, work, codex, browser, files, and actions. And if you know absolutely nothing about like, well, codex, right, or why does it matter, you know, why would, why do I need a desktop, you know, something running on my desktop versus something, you know, that I can just do in the cloud, right?

Starting point is 00:24:37 If chat chbt has this new chat gptv work on the web, why would I ever use codex? Well, the main reason is it can access your local system. So it can use your entire computer. It can use your browser, right? It can access read and write all of your local files. So those are, you know, three big advantages. And, you know, there's a little bit more that you can do with scheduling and things like that. If you have a machine like I do that just stays on, my poor Mac studio, I don't think is taking a nap since I got it.

Starting point is 00:25:10 Right. So I have a lot of things that are scheduled, you know, loops, goals, heartbeats, all these things. there's a little bit more technical capabilities that you can do with the desktop app. But it is now all one. There's some sinking issues, FYI, and I'll probably reach out to Open AI about it. I don't know if they're fully aware because as an example, some things change names, right? So if you were using Codex before, there was a thing called chats. Now there's no Codex chats.

Starting point is 00:25:43 Codex chats are now tasks. So they could be called codex tasks or they could be called chat GPT work tasks anyways. But you can, there's a new chats panel. And that brings in your chats from chat GPT. But right now, at least only chats from the chat tab inside chat GvT are sinking and not the chats from the work tab inside chat. GPD. Yeah, confusing. Right.

Starting point is 00:26:07 So there are some things that, you know, the, the super app and bringing all this together. in one cohesive ball. I think that there's a couple of things that the team is probably going to tighten up. But, I mean, first iteration, it is really, really good. All right. The other thing, small footnote, sites, all right? Chachapiti sites rolling out to all paid plans. All right.

Starting point is 00:26:32 I think there might be some geography restrictions at first. But rolling out to all paid plans. And this is important. Also, there's public publishing, live you are. RELs and it's on Chad GBT on the web, not just codex. So we are starting to see, you know, a little bit more feature parody, you know, certain like a good example is Chad GPD sites used to only be on Codex. So, well, now that's also on Chad GBT on the web as well. So here's why this is important.

Starting point is 00:27:05 Because before sites was number one only available on codex. So now it's available on the web as well. but it's also available for all paid plans. Before it was just for team plans. If I'm being honest, it was one of the reasons that I still was holding on to my team plan, even though I'm mainly the only person that uses it, was for chat, GPT sites.

Starting point is 00:27:30 I really like it, but on my $200 month pro plan, it's like I wanted to be able to use sites, now I can. So if you have a paid plan, chat should be these sites is great. It's like a lovable light. Right. So all, you know, I did a full episode on it a couple of weeks ago. Think of it like this. You know, think of how many versions of decks that you have, right? There's 10 different versions on 10 different people's machines, V1, V3, V4, final. Oh, did we have edits on this one? Right. You probably waste hundreds of hours a year just dealing with who has that file, right? Chatshavity sites can just get rid of that because it's just one source of truth. And anytime one person updates it, it's updated.

Starting point is 00:28:11 everywhere. So you can build full-on, you know, databases, dashboards, you know, little apps that your team can use or just decks and websites to keep track of anything type of work. So it's nice that it's rolling out to all plans and available on the web. Next, ChadGPT mobile got a huge update. Now you can manage plugins from ChadGPT app. You can see and manage all of your schedule tasks. So, That's great. But also, they changed the name. So you used to open up the chat GPT app and it would say codex. And then you could essentially control your any physical machine that was running codex. So now it's just called remote, which I kind of like, right. So they're not slapping the name codex everywhere and confusing people. And if you see remote, you're like, cool, even if you're a non-technical person. And now of a sudden you're like, wait, I can use the chat givt app, which, you know, opening eye has like a billion. weekly active users and everyone knows the chat gpti app and then they see this thing called remote and they're like oh i can you know control my desktop if i have the chat gpti work app wow great so again not really new

Starting point is 00:29:24 functionality there it's just cleaned up and it does work a little bit better but actually no new there is new functionality because you have more control over plugins and schedule tasks on the actual code app sorry uh on the actual remote section of the chat gbt app All right, number four, yeah, Atlas is technically gone. So opening eyes browser experiments. It is no more, but it's for the best because now this is going to sound small. All right. But two big updates to the browser formerly known as Atlas that is now just baked into the super app, right?

Starting point is 00:30:02 So Codex had always had a browser, but it was a singular browser. And it didn't have all the normal browser capabilities. Like you couldn't open multiple tabs. And it didn't keep your saved passwords. I actually spent like 1.8 billion tokens building what I called. I forgot what I called it. I think I just called it like Codex Vault. But I essentially built an app that would save all of my passwords so I could run, you know,

Starting point is 00:30:30 certain things agentically that required you to log in because, you know, the Codex browser, you know, couldn't log in to things without saved passwords. So that is available now. And you can import from any of your Chrome profiles because the browser is based on chromium like Atlas was. So it supports cookies to keep you logged in and it'll support passwords. So I was playing around with it a little bit because it was a little finicky to actually get computer used to click login because some websites.

Starting point is 00:31:06 depending on how they're set up, you know, the actual browser control can log you in. So it'll pull up your saved, you know, websites, your saved credentials, right? Just like in Chrome. If you, you know, sign into something and you've saved your password in Chrome, it'll be there and you just have to click log in. So I had to toy around with it a little bit. I might just, if I can refine it, I just might save it as a skill and, you know, put it in our newsletter or on our website because it is a little finicky now.

Starting point is 00:31:34 I'm sure they'll get it fixed. But this is the big unlock here, y'all. And here's why. Yes, you know, there's connectors and plugins and MCPs and all these things, right? But I think for knowledge work, so much of what we do is going to all of these different websites that don't have MCP and they don't have direct connectors, right? So in my example, you know, our podcast is hosted on BuzzSpout, which is an older host. They don't have any AI anything.

Starting point is 00:32:03 I spend so much time, you know, going from this program, you know, downloading this, uploading this file here. I clean the audio. Then I bring it in, you know, get the timestamps. And it's like five different programs. Well, now I can just fully automate that completely because codex can open multiple tabs at once and it can save the passwords because, you know, it does save them, had previously saved them for a certain amount of time.

Starting point is 00:32:31 But every once in a while, it just kicks you out. like any normal, you know, SaaS program would. So now with the cookies and the saved passwords, huge in terms of the new knowledge work that you can do on a schedule. Think of all those monotonous tasks. You have to go log in, click this thing 12 times. You're like, why am I doing this, right?

Starting point is 00:32:52 Why, you know, even if you're a high paid executive, you're like, I, you know, I just every single day, I have 45 minutes of things that, you know, a monkey could do, right? Just go here, click this, wait for it to load. three minutes, oops, got distracted, click this, export, download, right? You don't have to do any of that anymore.

Starting point is 00:33:09 It's amazing. All right. And then last but not least, computer use is much faster. It is faster. It is better. And it has this real cool new thing. It's this picture and picture.

Starting point is 00:33:20 So when computer use is happening, you can see, you know, if you've ever seen picture and picture on your TV, it's just right there in codex. Or if you decide to call your codex chat, GPT work. So you can see what's happening right there in Codex. Oh, another thing that I didn't mention, which I should have mentioned on number four with Atlas, related to this and related to this super app. And I don't even think OpenAI announced this. And I didn't see anyone else talk about it,

Starting point is 00:33:51 but in the Chrome extension, because you can also from Codex, not just control your computer, not just control the built-in browser, but you can also control a Chrome browser, which is a little different. All right. But they actually have a Chrome extension. And there's an update in the Chrome extension where you can have a side panel in Chrome and you can see what is being worked on on your desktop, which is really cool. So, you know, if I have a window open, which is actually unlocks a lot because normally on my setup, right, I have two screens. One screen is just full, you know, full codex, you know, and then I'll have, you know, probably a full Claude desktop. And then I have to swap between them because I'm like, oh, okay.

Starting point is 00:34:36 And I know, you know, Codex has the little pets and they say, all right, your thing's ready. But I'm still like having to always keep an eye on that. So having a sidebar in Chrome and I can see what Codex is working on and track it live. That's been like a small little thing. And I'm like, wait, why didn't they talk about this? This is huge, especially if you're like me and you're giving, you know, agents long running tasks. You know, at any point I might have, you know, five to. 20 different threads running.

Starting point is 00:35:05 And it's like, I don't want to always have to check in on them. And sometimes the little Codex pet for me is a little distracting. And I'm always in a Chrome browser. So I'm like, oh, okay, I can just have that going and work with it right there. Actually amazing. All right. So that is the five overlooked features. And let me end with this hot take.

Starting point is 00:35:28 For those of you that stuck around, Anthropics in trouble. they are in trouble. If they do not immediately pivot, it's not going to look good. All right. And here's the thing. Yeah, Anthropic is scared. And they should be, right?

Starting point is 00:35:50 And this is great because it's going to drive competition. Anthropic did something that I don't think they've ever done right after this announcement. And this is how I know Anthropic is in huge, huge trouble. they reset weekly limits, which I don't think they've ever done, except one time when they made a big mistake. There's this big miscommunication. I think they might have reset five hour limits, not weekly limits. That's something if you use codex, like we've been spoiled, right? Tebow over there at OpenAI, always resetting your weekly limits.

Starting point is 00:36:27 Infrappit did this. They never do it. And they did it right when Open AI announced something. And that to me is like they know they're scared and they know I'm not going to make projections. I'll make a projection. I'd say it's very, very safe to say that yesterday they lost probably eight figures in revenue, I would guess.

Starting point is 00:36:48 And I will continue to say that until they make a change, they're going to be losing eight figures of revenue every single day, maybe more. They're going to be losing a lot of money if they do not pivot immediately. So I even heard some people say, and I'm not going to say this, right? People are like, oh, Fable 5 is dead. It's not dead. It's still, you know, a top tier model. And there's still going to be some people out there paying for it.

Starting point is 00:37:11 But I would absolutely not pay a dime for Fable 5. And Anthropic, they said, oh, we're going to, you know, extend Fable 5 in subscriptions through July 12. Well, if they don't extend it past that, they're going to lose subscribers. but the majority of Anthropics business comes from two main things. It comes from coding via the API, right? And up until 5-6-Soul, they had a sizable lead, right? Fable 5 had a sizable lead over 5-5 from GVT.

Starting point is 00:37:45 And, you know, Google, we'll see what happens. But here's the thing. I don't think Anthropic is worried about GPD-5-6-S-Sou. They are worried about JPD-5-6-Soule. about GPD 56 terra. Okay, here's why. So, Anthropic makes money from coding.

Starting point is 00:38:07 Yes, they have co-work. Yes, that's popular. They make their money from coding. They sell tokens at an extreme premium. All right. Looking at the artificial analysis coding index,

Starting point is 00:38:20 and that takes into account a lot of the popular coding benchmarks, you know, So, GPD 56, soul is in first place. But guess who's in second place? GPT-56 Terra, the medium model. It is better than the big, bad fable. The one that we had, you know, have a government anthropic fight, the one that, you know, Anthropic had been, you know, saying, this thing's too dangerous.

Starting point is 00:38:46 They've been, they were hyping it up for three, three months. And now we have this, you know, permission slip AI in the U.S. because of anthropic over hyping Fable and mythos. Okay. The medium model is eating Fable 5's lunch. And why are they eating it not just because it's incrementally better at coding? It is more than five times cheaper according to artificial analysis. Five times cheaper to do the same tasks versus.

Starting point is 00:39:25 Fable 5. Yeah. So what I said on the Twitter machine, I said, Anthropic has like a week or two until enterprises start catching up. Because that's how long it takes for even the decision makers

Starting point is 00:39:38 that are spending multiple millions of dollars, sometimes a month on these products. And yes, there's vendor lock-in. But any company that's up for, you know, if they signed a one-year deal with Anthropic, there's always companies rolling. They're going to churn. There is zero, zero reason for any company right now to be paying for Anthropic.

Starting point is 00:40:02 Zero because of what OpenEI just did. They undercut on price. They overdelivered on performance. And if Fable, if Anthropic doesn't change Fable 5 pricing, it is DOA. It's done. So they've got to either quickly release Fable 5.1. so they can, you know, knock down the pricing on Fable 5, or I do think that they will be bleeding five figures potentially daily, right?

Starting point is 00:40:31 So not just from people going over on contract renewals, but just on the API side. It's going to be interesting to look at these third-party sites, like OpenRouter that track, you know, third-party token usage, because I would assume you're going to see the Open AI share go up exponentially, and you're going to see people that we're using, at least for Fable, go down. Opus, go down, because why would you use Opus now when Terra is cheaper? So a big portion of Anthropics revenue just got caught off at the knees.

Starting point is 00:41:06 And if nothing else, this is just fun for us, right? We get to watch, but that means we're going to get, number one, a better model from Anthropic sooner rather than later. They have to. And they're going to have to cut their prices. or extend and keep Fable 5 in subscriptions. There's really no other option, right? Because they can't just keep bleeding money

Starting point is 00:41:29 because they're going to be bleeding money. All right. So that's it. Now you know chat GBT work. You know GPD 56 soul. What's new? The five overlooked features and my one hot take. I hope this was helpful.

Starting point is 00:41:40 If so, please go to your everyday AI.com. Sign up for the free daily newsletter. Thanks. We'll see you. Well, not tomorrow, but Monday and every day. More Everyday AI. Thanks y'all. Here's the thing about AI at work. You can't use what you don't trust.

Starting point is 00:42:01 The all-new Slackbot runs inside Slack's security boundary only sees what you've already allowed it to see and never trains on your data. Personal AI agent with the trust to actually put it to work. Slack bot from Slack. Learn more at Slack.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going.

Starting point is 00:42:33 For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.

Everyday AI Podcast – An AI and ChatGPT Podcast - Ep 816: ChatGPT Work and GPT-5.6 Sol: What’s New, 5 Overlooked Features and 1 Hot Take

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.