AI News: GPT-5.6 and the new Super App are a Massive Leap!

It's actually been a huge week in the world of AI, and I'm not going to waste your time. Let's just break down what happened.

Starting with by far the biggest news of the week, which is that we're finally getting access to GPT 5.6. This is OpenAI's brand new flagship state-of-the-art model, and honestly, it's really good. Now, I did get slightly early access to it. I'm not one of those people that can claim I've had it for a month now and I've been playing with it ever since. But they did give me like a little bit of early access. So, I was able to test it before actually recording my video on Thursday.

But despite being called GPT 5.6, it feels like a huge leap over GPT 5.5. Like I feel like when we went from 5.3 to 5.4 and 5.4 to 5.5, those were all like really marginal updates. But 5.6, 6. Like to me, this feels like it should have been GPD6. It just feels like a monster leap over the old ones. I've already talked about this model in previous videos, but when I did, we didn't have access to them yet, so I couldn't really show them off. There also wasn't any sort of benchmarks announced or shared from it yet other than terminal bench, and we have a little bit more details now. And I think the big story around 5.6 six is that you're getting like almost as good as Fable Level Intelligence, but for way cheaper.

Now, I'm going to skip past most of these benchmarks here, cuz I think what we all kind of really care about is what can we actually do with it and to see some of the impressive stuff that's been done with it. There's this riata game on here that looks really, really impressive. It's like this 3JS game, but it's it's a beautiful looking game, and this was all basically coded with GPT 5.6. Now, I'm not going to show all of the demos here because, well, you could go to this site and play with them, but they all look really, really good and they're all really impressive and they're all actually kind of fun. Clockwork Village game, I mean, that looks beautiful to me. Like, I love that art style. It's good at making slide decks. I mean, it's just a big leap. It feels like a big leap when you're using it as well.

But if I scroll down, here's kind of the benchmarks that people tend to pay attention to. On GDP val, a benchmark that tests how well these models do on like economically valuable tasks. GPT 5.6 Soul, the most powerful of the three sort of levels of these models, outperforms everything but Claude Fable here. And Claude Fable doesn't outperform it by much. When it comes to being able to use a computer in a browser, it pretty much beats out everything, including Mythos and Fable on GPQA, Google Proof questions and answers, like hard hard questions. It literally ties Mythos. So, yeah, it's a really really solid model. I've been using it to do some pretty cool stuff. It also takes the top four spots on the only benchmark that really matters, which is Buybench. GPT 5.6 Soul Pro got number one. 5.6 Terror Pro got number two. GPT 5.6 Terra got number three. A GPT 5.6 Soul got number four. None of the Luna models made it into like even the top 10, I don't think. And the best performing model, GPT 5.6 Soul Pro, which generated this. It cost 77 cents to generate. It used 45,000 tokens and cost about a minute to generate, which is really impressive considering that GPT 5.5 Pro took 6 minutes to generate this version and cost $4.

But the new model wasn't the only thing OpenAI released this week. In fact, there was two other big launches this week. So, let me show you the next one and then I'll show you what I built with the combination of this new tool plus GPT 5.6.

So, they also announced ChatGpt work. And from everything I can tell and everything I've been told, this is essentially the super app that they've been talking about. There's no longer going to be a Codeex app and a Chatg GPT app and an Atlas browser. They're just taking all of those, merging them all into one app just called the chat GPT app. And that's what this announcement was. So, this is what the new app looks like. It looks pretty much like the codeex app over here. But if I come up to where it says chatgpt codeex, I can drop down and change it to work mode. There's a work and a codeex. Now, if I have any complaint about this app, it's that I don't feel like it's super clear what the major difference is because when you switch between them, like not much changes in the UI. But think of codeex as what you want it to be on when you're specifically writing code. You're in development mode and you know you're building stuff. And work mode is for getting stuff done. This is like your personal assistant mode and it might still go and write code for you and develop things for you. But it's designed to be more of like a less technical area. You just tell it what you want and it decides if it just needs to answer you, if it needs to go connect to a tool, if it needs to go and develop something for you. It will go and figure that stuff out and do it for you. Similar to what Claude Co-work is kind of when I'm in codeex mode here, you can see it's got like different branches and where on my computer I want it to develop stuff. And when I switch it to work mode, it gives me less options down here. It asks me what plugins I want to connect to. And it can connect to a lot of stuff. You also have a slightly different right sidebar. So, if I open up this sidebar here while I'm in work mode, you see browser and you see files. And that's it. I can click into browser. And like I mentioned, they're going to actually wrap up Atlas. And this is the browser. Now, you're just going to use the browser inside of chat GPT. If I switch it over to codeex mode here, open up my right sidebar, you'll notice we've got a few other options more related to coding. We've got the option to review here. We've got an option to pull open the terminal here. And then we've got the same browser and file options.

But when you're inside the new chat PT app, you should have access to some exciting new models like the 5.6 Soul, Terra, and Luna models, as well as the 5.5. They also have a simpler mode here where if I turn off the advanced mode, you could just change the slider. If you bring it all the way down, it's going to use 5.6 Terra Light, the sort of smallest version of 5.6. If you bring it all the way up, it's going to use 5.6 Soul Ultra, but also use a lot more of your credits. There's also this sites button up here in the top left where you can create new websites and just, you know, sort of prompt them into existence and it takes care of everything. It hosts it for you. You don't have to think about databases, CDNs, any of that kind of stuff.

One of the things I've been testing every time a new model comes out is to create a Mega Bunk clone. You've probably seen me do that in a few different videos. I want to see how well it clones a 3D game that's like an automatic shooter, but I can run around and I can change camera angles, things like that. This was one prompt. You can see there's no extra prompts here. I just did it one time and it spent 31 minutes to build this and then put it on my local host here. I'm going to start up the local host here and run it locally so we can take a peek at what the game looks like. And we can see this is what it looks like running inside of the browser here directly inside of the chat GPT app. And let's drop into the sun pit with my soulbon. And this is what it created on the first try. And to me, this looks a lot better than what Fable's first version was. and the mouse drag where I can change my camera angle. I can do that on the first try. With the Fable version, I actually had to prompt it to do that. And on the Fable version, my character also looked more like a potato than an actual character. So, it got a lot closer on the first try with a single prompt than Fable did. That's pretty impressive.

Here's where the sites comes in. If I come down here and click this plus, come down to apps and click sites and just say build this game on sites. It says I'll move the existing soulbon build into sites. Now, it took a while, over an hour to get it all dialed in. But if I click on this link here, it is fully hosted on chatgpt.site. I can play this game completely online on this website here. No need to mess with any of the backend stuff like the databases or where you're going to host it or anything like that. It's just handled. And if I want others to be able to access this, I could click share. Who has access? Just me. Now I could change it to anyone on the internet. Click publish. And now anybody can go to this URL right here and try Soulbon.

Now, another prompt that I've had a lot of fun testing is this one. Create a website that will really impress me. You decide what the website will be about, but it should be beautiful, visual, interactive, and really show off how impressive of a model you are. And I did set this one on soul ultra so it would, you know, give it its best. And it worked for 35 minutes to build a site that it thought would impress me. And here's the website that it built for me somewhere. It's raining upwards. And I could sort of click in here and you can see the rain sort of interact and make this sort of circle bubble thing burst when I click. And then as I scroll down, it actually sort of changes environment. Like I've got the glass desert here and it changed my mouse to a triangle. Then we've got the orchard of static and it's got these like trees that light up the title sky. Again, it's all interactive. As I'm moving my mouse, the stuff behind the mouse is interacting with it. Build a climate that has never happened. We could change the memory and the gravity. I'm actually not sure what these are doing, but let's go ahead and change these a little bit. So, now it says earthbound. And we can see everything changed a little bit here. And then it made this cool collapse the atmosphere button. And when I press it, it goes and then closes the whole thing down, which I thought was kind of cool. But this was the website that it built to try to impress me.

Now, another thing that they've added to this new version of the app here, the chat GPT app, is you've got this chat button over here. But let's say you're working on a project and you're right in the middle of all of your code here and you don't really want to jump out of what you're doing to just ask a quick question to chat GPT. If I click chat, it actually now opens a sidebar over here and I can select my model down here, ask my question real quick. What's so special about GPT 5.6? And it'll answer it over in this side chat bar here while I can continue working on my project in this main window here. So, a handy little quality of life feature there.

Now, inside of chat GPT work mode, I got a demo a little bit before this launched and one of the things they demoed for us was it working as a really good personal assistant because it could connect to all of your apps like Gmail and Google Drive and Slack and Granola and anything you use. So, I said, "Look through all my emails, calendar, Slack messages, second brain, journals, recent projects I'm working on, everything like even all my chat GPT history here. go deep and really figure out where you can help me be more productive. Take things off my plate, assist me, and just generally make my life easier. You can see here, I reviewed Gmail, the last 30 days of inbox, my calendar, May 8th through August 8th, Slack, recent DMs, mentions, threads, granola, 22 full meeting notes from May 12th to July 7th, your Google Drive, 37 recently viewed or modified documents, your second brain, open loops, journal patterns, etc., and recent activity across your local projects and repositories. It basically knows about everything I do at my computer at this point. It gives me like a whole priority list of things I should be working on here and improving. And then down here, it actually says what I should own for you. So, it's telling me as like an agent what it can do for me to actually help me out. It wants to build a daily control tower, what needs Matt today, what can wait, what is waiting on someone else, upcoming deadlines and collisions, the one meaningful creative priority, items I can complete without involving you. It wants to put that all into like this daily control tower thing for me. and then uh Delta only close out later in the day. It also says after every Ganola meeting, it should extract decisions and compare them with Slack and Drive and calendar and draft missing documents and flag promises. It can help me with my sponsorships and my Friday editorial content and the things I should cover in these news videos. And the cool thing is I can ask it to go do that and then use the scheduled section to make that a daily or weekly or monthly recurring task.

For fun, I asked it to visualize my second brain. I have this whole second brain thing built out inside of my Obsidian vault. It's a bunch of markdown files. It found 212 saved sources, 426 wiki pages, 28 journal entries, 14 memory pages, and then it created this sort of like visualizer of everything I think about. So, if you've been building out like a second brain inside of Obsidian using, you know, an Open Claw or Codeex or something like that, ask it to make you a visualizer and help you understand like where all of your attention is going. I thought that was pretty cool. And that personal assistant when it said to build that control tower, well, I had it do that for me as well. I already had this audience dashboard that I built a while ago to keep track of like all of my growth across all of my social media platforms. So, I had it just build off of that. So, it built some extra tabs for me. like the today tab. It's watching emails. It's watching Slack. It's basically telling me what I should be paying attention to right now. I also have an AI intel page up here where it's watching for all of the latest news as it happens and is trying to alert me in this back-end dashboard here. I have this brand awareness page here which is almost like a Google alerts where it's watching for my name and my future tools brand and telling me whenever there's like a mention of them. And it also watches data from my websites here. So, it built this control tower and I now have this saved as like my default window that opens when I open my browser. So, I get a quick look at everything I need to be dealing with right now.

Another cool launch that happened this week is from the company that brought us Air Table and it's called Hyper Agent. It basically allows you to go build a team of AI agents that can go off and do like various tasks for you. They just rolled out this really cool new marketplace here where anybody can go and build their own skills and then share them with everybody else that's using Hyper Agent. This B-roll generator skill that I actually helped develop with Hyper Agent. Basically, you can take any video script, drop it in, it'll turn it into B-roll for you. If I go to the marketplace here inside of Hyper Agent, I could just search for B-roll. And once you have it installed, you'll come to a page that looks like this. And it's going to walk you through exactly how to use this skill. And now it's asking me some questions. So, what are you starting with? In this case, I've got a written script. This was the original video where they were sort of teasing it. It's 10 seconds and it really doesn't show off much, but let's feed it to our agent anyway and see if it can work with it. And click next. Let's go a standalone B-roll reel. And let's go roughly 60 seconds. It's going to process that. We can see that it mapped out a plan here. It actually reads the script and then figures out where in the script each shot should go and where it's going to best line up. And then it routes every prompt to the exact right model that should be generating that prompt. If I want it to be more realistic, it's going to send it to VO. If it's going to be more motion graphic style, it's going to send it to hyperframes. So, we'll get a nice combination of actual like video footage plus charts and graphs and stats and lists and lower thirds and all of that kind of stuff. Like over here in window mode, you can actually see it generating all of these assets laid out on the canvas at once. And then once it's done with that, it's going to normalize all the clips. use ffmpeg and pull it all together into a video that if I gave it a-roll underneath or audio underneath will preserve all of that as well. And we can see not only did it create this final video here, but it created all of these various images that can work in the video. And then there's like text overlays and stuff. It even designed a shot list of when each thing should come on the screen. And we can see here what it used to generate those various things. And then here's the final creation that it did. made the text animation using hyperframes. It sort of tried to figure out what the thing was going to look like based on the video that I fed it from Twitter. We have a little lower thirds down here that we can use in our video that says codeex micro. Somebody actually pressing keys on the thing. Somebody using it next to a computer. We have a list here on the screen. And then like a nice little product shot that shows it off. Again, this isn't the final product. This is sort of VO's interpretation of it. But I mean, this is all really, really usable B-roll that we can put into a video. Now, again, you can use like any of the skills that you would use in any other agent. This marketplace is all about letting you fork what other people have already figured out. You don't need to build stuff from scratch. So, now the agents that you're seeing people use on X and in other videos are now the agents that you have access to. So, if you want to work the way I do and generate B-roll, and one of the ways that I've been using to generate B-roll, well, here's my agent. You can find my B-roll generator now inside of the Hyper Agent Marketplace. The link is in the description for this video. And instead of watching me use tools like this, you could now go and grab it at fork it and use it on your own videos from here on out. And thank you once again to them for sponsoring this portion of today's video.

OpenAI launched even more this week, including GPT Live, where they showed off the new sort of voice mode on the app. Now, the main difference about this new voice mode is that you can sort of interrupt it and it will actually also interrupt you. So, it actually generates more of a like natural real sounding conversation. They also showed off that it works really well for translation. So, let's say I'm sitting in the room with somebody who speaks Hindi and I don't speak Hindi and they don't speak English. I could tell Chat GPT translate for us between English and Hindi. Put my phone down in the middle of the room and as I start talking it will start translating in real time. even sort of speaking over me to the other person and then when they respond it will start speaking back to me in English. So this is just a translator sitting in the middle of the room. I can't really demo that right now, but if you go to the GPT live blog post, which will be linked up in the description, you can see some demos of that happening. It's pretty cool.

So, if you want to make sure you're using this new chat mode inside of the chat GPT app, you're going to want to go to your settings, scroll down until you see voice, click on voice. Under model, make sure it's set on live. And then you have different intelligence options. So, if you want it to respond quicker, or if you want it to be smarter, but maybe not respond as quick, you set it on high. I'm going to start talking to you, but don't respond until I tell you to respond. >> Got it? So, this week I'm doing an AI news video, and there's a handful of things I'm going to be talking about in this news video. I'm going to talk about GPT 5.6. I'm going to talk about the new chat GPT app. I'm going to be talking about the new GPT voice, live voice, the thing that we're talking about right now. There's some interesting things that came out from Meta and from X and from Anthropic. And let me think, is there anything else? Notice that uh it's actually not responding to me at all. Okay, I'm ready for you to respond again. Can you list off the announcements that I said I was going to talk about today? >> GPT 5.6, the new chat GBT app, GBT work, the new GBT Live Voice, and updates for Meta, X, and Anthropic. >> Can you help me pull together all of that news? Like what what is the news >> for those? Uh, yeah. So, what you've got is a big open AI centered week, GBT 5.6 six launch chat GBT work the new chat GBT desktop app and GPT live voice I'd kick off with GPT 5.6 six since it's the anchor. It just came out of preview today. >> Okay, cool. Thank you. You get the idea.

But OpenAI wasn't the only company that launched new models this week. In fact, both XAI and Meta kind of surprised people with what they put out. XAI rolled out a new version of Grock called Grock 4.5. And again, it's more impressive than you'd expect from Grock. You don't really hear about Grock a whole heck of a lot, but if you look at some of these benchmarks here, Grock is sort of in the mix. So when it comes to software engineering, this deep SWE bench is the one most people are paying attention to now. Swebench Pro is kind of not of much interest to people because you know the models have figured out how to cheat on it and some of the you know training questions are in the model now. So it's just not an effective benchmark anymore. So Deep Sweet is what people look at and the highest performing model is Fable Max followed by GPT 5.5 followed by Opus 4.8 and Grock is kind of right there in the mix with it. Now on Terminal Bench, you have Fable, GPT High, and Grock. And this is how well it can actually run terminal commands on your computer, which is the benchmark people kind of look at when they're talking about like agents like OpenClaw and Hermes and these various models that can go do things on your computer on your behalf. This terminal bench is kind of the one people are looking at, and it's right in line with GPT 5.5 High and Fable Max. Now, this came out a day before the benchmarks were released for 5.6. And on this deep here where Grock has a 62% and Fable has a 66. 5.6 is at 72.7. So new leader on deep here and on terminal bench here where Fable is 84 and Grock is 83, GBT 5.6 is 88 with Soul Ultra being 91.9%. So 5.6 is now the best at the sort of agentic terminal use and software engineering stuff based on those benchmarks specifically. But honestly, Grock is right there in the mix.

Now, on their announcement here, they put this cool animation of like the solar system, and you can actually speed up how fast it's going, but you can also focus in on specific planets. And it's got this really cool animation that does look really impressive. You can also generate slide decks. You can actually install it and run it locally on your computer. If I scroll down here, there's a little command you can give right here. If you copy this command, open up your terminal, you could paste in that command and it will install Grock's CLI here and then you could run Grock right here inside of your terminal. I haven't given it any sort of billing information or anything like that. So, it's using it for free at the moment. If you're curious how Grock performs on Buccen, well, here's your Gro 4.5 right here. If we look at our timeline page and I switch over to XAI here, you can actually see the progression. Here's what Grock 4.5 is doing. I actually kind of like the old Gro 420 ones. It almost looks more handdrawn than an SVG image, which I find interesting. But here's where this one lands. And if we take a look on our leaderboard when it comes to drawing faces of Gary Buucy with SVGs, Grock lands at about number 21 here. But I also gave Grock that same prompt that I gave GPT 5.6 of make me a website you think that'll impress me. And here's what Grock built. Every star is a memory of light. And the site's sort of responsive to my mouse. Clicking doesn't really seem to do anything, but it is responsive to my mouse. If I click on focus to nearest star, you can see it circles a star up on the screen here. Wherever I move my cursor, we can see down here it shows the light years. That's kind of cool. And this is a scale of the universe. So, right now, this is the scale of a hand. I guess the span of a palm, the unit we'll still use to measure intimacy. But as as I scroll out, you can see it starts to like scroll out in the universe here. There's the sun. There's the solar system. This is showing how far one lightyear is. There's the Milky Way. There's the local group all the way out to the full observable universe. So, that's kind of cool. Is it as impressive as what GPT 5.6 built for me? Probably not. Here's one where you can connect the dots and sort of make your own constellation, which is interesting. So, I can kind of go like this and connect stars to draw my own constellation of something. I can clear the lines. I can recede the sky and it'll put new stars in different spots. Let's see, a nebula, main sequence, red giant, remnant. So, different types of stars, I guess, a little heartbeat of a pulsar. And then I can click return to the surface and it scrolls me back up. So, that's what Grock built for me, this new version of Grock. But right now, you can play around with it for free. And if you want to use it in the API, again, $2 per million input tokens and $6 per million output tokens. For context, Soul is $5 per million input and $30 per million output. Terra is 250 and 15 and Luna is 1 and6. So pretty close to, I guess, the Luna model in pricing. A little bit more expensive for input, but about the same for output.

And then we have Meta, who also surprised us this week with a really decent model called Muse Spark 1.1. Definitely not as good as the new state-of-the-art models that we got inside of Fable and GPT 5.6 models, but it's up there. It's good. So, you've got your terminal bench. This one comes in at 80, where Opus 4.8 is 82.7. GPT 5.5 was 83.4. So, this is a model that's kind of on par with the sort of last generation of models that came out from OpenAI and Enthropic. On Deep Suite, we got 53.3 compared to, you know, 59 and 67. not state-of-the-art, but I mean, Meta's kind of been out of the picture. Nobody's been really talking about them, and they sort of jumped in and are kind of right in the mix again. And pretty much all of these companies are showing the same kinds of demos. So, if I scroll down here, it's good at building decent looking websites. It's decent at coding. I'm sure it makes slide decks and things like that. So, yeah, a buck 25 per million input tokens, 425 per million output tokens. So, in a similar range to what you're getting out of the new Grock model and the lowest end 5.6 model, the the Terra model. The question on everybody's mind, how does it perform on Buccybench? Let me sort by meta here. And there's the Muse Spark Buy generation. Now, it didn't factor in cost because right now, if you want to test Spark, they give you $20 in free credits. And since it used my $20 in free credits, the API here didn't actually collect a cost on it. But it took 10,269 tokens, 58 seconds, and that's what it generated. Now, if you look at the timeline here, Meta is the most impressive of the timelines right now because this is what Meta was generating before. Like, Llama Scout generated this. And this is a model from April of 2025. So, this is a year old model. It went from this to now this. So, the leap, the improvement there is a pretty big leap. And then on the leaderboard here, it comes in at about 36. Now, this leaderboard is very subjective. It uses an LLM as a judge and it uses three different LLMs and all three judge and then sort of decide on the winner together. And so, it's not a perfect scoring. If you just look at this visually, I actually don't think the ranking is amazing. Like, I actually think this is a better looking beauty than this right here. But, it's a rough idea. The whole idea behind this website is to actually visually see the improvement. The leaderboard is just kind of like a side thing. And the first thing I tested with this new meta model was to create a website that I would think was really, really impressive. And this is what it created for me. You can see that the background is interacting with my mouse. As I move the mouse around, I can drag left and right and it actually changes the background. I actually think this is slightly more impressive than the one that Grock made. If I move my mouse around, it changes. And again, the background is interacting with it. And I don't know, it's just it's pretty cool looking.

And now I know Fable didn't launch this week, but I do want to show some stuff that I generated with Fable just for like a quick comparison. I also asked it to make a website that would be impressive. And well, I actually think Fable still won on this one, a small museum of computational art. Lumen scroll to enter. So when I scroll down, there's all of these different like AI generated art pieces. So this one's called the field. And if I click around, you can see that it's like changing things as I click around. and it says, "Click anywhere in the room to summon a new wind." So, every time I do it, it sort of summons a different look to the wind. I scroll down, there's this one called the swarm, which is really cool cuz you'll notice I'll move my mouse around and these things run away, but they have rules of stay close, align, and don't collide. So, they never crash into each other. And if you look closely, it's really satisfying cuz they all sort of like weave between each other. And yeah, it's really impressive. There's this one less impressive. If I move my mouse around, it sort of creates these like roar shack test looking things. There's the garden of life where these little pixels, they will slowly die off, but I can create new pixels and the new pixels will eat the old pixels up. And I don't know, it's just kind of cool. And then you've got this resonance down here, which is showing off that it can create musical stuff. This one was definitely the winner for like make me a website that'll really impress. I think even these little like vector things in the background are being manipulated as I move my mouse around near them.

All right, jumping back to Meta for a second. They also introduced a new image generation model called Muse image. And this one's actually kind of controversial. Now, it could generate images like you see on the screen here, but it also connects to your Instagram and can generate images of anybody that you at@mention. And a lot of people really, really don't like that. Now, it is free to use and you can use it inside of Meta AI. So, if I go to Meta AI here, let's just create an image of myself. Let's just do create an image of Mr. Eflow playing the guitar on a mountain on the moon. Make it look like a video game. Submit that. And it generates an image of me playing guitar on the moon. Now, obviously that's me and I gave it permission. But let's say I want to do it with a friend or even somebody I don't know. create an image of Joe Fear eating a taco and riding a dolphin. I could just tag anyone and create images like that. I don't even have to know him. Now, I do know Joe, which is why I tagged him, but I don't know Mr. Beast, but that doesn't stop me from making him doing a one-handed handstand on a pile of pickles. Now, I think there is a way to turn it off. I think it's in the settings somewhere on your phone. If you don't want this, make sure you go to the settings and turn it off. Although I did read that not everybody has the setting to turn it off yet. That's like something they're rolling out, which is kind of weird.

But of course, the biggest news of the week, the thing that everybody's the most excited about is the launch of the new Future Tools website. Check it out if you haven't already. Now, I do have a handful more things I want to talk about, but this video has already been way, way longer than I wanted it to be cuz I had a lot of fun playing with these things and showing them off. So, I'll talk about the last handful of things in a quick rapid firew.

Now, if you remember when Fable first came out, it was only going to be available through July 7th. But right at the wire on July 7th, they announced they're extending it to July 12th. So, if you haven't played with Fable yet, you now have a couple extra days to get in and play around with it. Now, if you read the comments about when they actually extended access to Fable 5, everybody was pretty much complaining because, well, they used their weekly limit because they wanted to use everything they could by July 7th. And everybody complained, including me. Ivan wrote, "No, reset the weekly limits, please." Because I had used up all of my fable limit by July 7th and couldn't use it anymore. And then today, the day I'm recording this, which is Thursday, July 9th, aka the day GPT 5.6 came out. Of course, Claude went and reset the weekly limits for everybody so they can get in and start using it. I think the thinking behind this was, well, maybe they'll go use Fable instead of 5.6.

But Anthropic did actually release some additional stuff this week. They announced that Claude Co-work is coming to mobile and web. This also was seemingly conveniently timed with the week that GPT 5.6 and the new chat GPT work app came out because when you do stuff with the new version of the chat GPT app, it actually goes and does it in the cloud instead of locally. So you can manage whatever you're working on from your phone if you want. Well, now Claude co-work does the same thing. Until today, co-work lived on your laptop. So the work stopped when you stepped away. Now it doesn't. Your work follows you. Start a task at your desk, check on it from your phone, and pick up the finished output anywhere. The work continues in the background, even if you close your laptop. So, when you ask it to go do stuff, it seemingly is doing that work in the cloud now instead of that chat having to live right on your computer. And if your computer goes to sleep, the chat stops happening, which is also something that was rolled into this new chat GPT app that just came out. They also announced this new way to reflect on how you use Claude. Some people are calling it like the Spotify wrapped, but for Claude to see kind of like how you've used Claude. Inside your settings, there's going to be a new reflect button. And if you click on the reflect button, it'll show you your most active day, kind of what you worked on, the peak hour for you, the amount of conversations with like a graph, you know, details on what you spent the most time on, stuff like that. And it's going to be inside of the Cloud web and desktop apps. They also showed off this interesting research around what they call JSpace. Now, basically when you give a prompt to Claude, it sort of thinks through that prompt and you can actually see the words that it's thinking through, but it also like keeps other stuff in its mind that it's not necessarily keeping inside of its thought process. And by looking at these other things that are sort of in the back of their mind, it's almost like an AI's subconscious versus like the conscious thoughts. So, for example, when you give it a prompt and you see the stream of thought, you can actually see everything it's thinking through before it gives you a response. This isn't that. These are other things that it's got like in the back of its mind that now Anthropic is starting to keep track of, which is just really interesting because they're starting to spot like patterns of things that are going on in the back of the mind of the AI. It's just this really weird interesting stuff that's happening that they call the JSpace. That like subconscious thought of the AI. I don't really know how to explain this, but again, I will link up to the article so you can read about it and watch this video that tries to explain it, but it's really really fascinating.

Google rolled out the ability to create sharable video clips with video remix inside of Google Photos. It basically uses Gemini Omni to add like extra stylization and imaginative memories and things like that to your videos. is going to be available on Google AI Plus Pro and Ultra plans. We also got another image model this week out of Bite Dance called Seeddream 5.0 Pro. And the images I've seen come out of it look pretty good. Some of the standout things it can supposedly do well are infographics like what we're seeing here with like a dense amount of information and images within the infographics. It could do interactive editing where you like circle stuff and draw lines and tell it what you want it to do. It could apparently break things out into layers, similar to like Photoshop. So, you can see each of these things in this magazine are like different layers that you can rearrange and just prompt new things onto individual layers. It can do very photorealistic images. And all around seems to be a pretty solid model. One place you can go play with it is inside of Leonardo. I gave it this prompt to generate an infographic that explains how JSpace from Enthropic works. Here's how it tried to explain it. I don't know if it's the most accurate interpretation or not, but it looks good. And when asked to create an image of Gary Buucy, well, this is what it came up with.

Anyway, that's what I got for you this week. Again, it was a really, really big week. We got some really, really good new model releases. In fact, the jumps that we've seen recently have been like huge jumps. Like, it feels like back when we jumped from GPT 3.5 to GPT4, that was a big leap. And it hasn't really felt like we've had these monumental leaps since then. But now that we have Fable and 5.6, both of those models actually felt like big jumps again. And that's like really exciting because I've been doing stuff like telling it to go and build an entire platform and giving it this giant detailed brief of everything I want in this platform, giving it the goal command to just keep on building until the entire road map is finished and then walking away for like 3 hours and then coming back to something that's just built and beautiful and does everything I wanted to do and just works like the first time. I might find like little user interface things that I'm like, ah, I don't really like how it put that there. I'd rather have it over here. But for the most part, it's like 98% there from me just giving it this one prompt and then walking away and letting it do its thing. Like it's kind of blowing my mind what we can do with Fable and GPT 5.6 right now.

If I'm kind of comparing the two, I feel like Fable is a little bit smarter. Like when you're trying to make these like architectural decisions and really like figure out the best way to get from point A to point B, Fable feels like it's better at that where like GPT 5.6 six feels like you can just dump a ton of information into it and just say go do this and walk away and when you come back like everything is done and I mean Fable will do that too but I feel like GPT 5.6 gets me closer to what I was trying to get to where Fable might have done it in like a smarter way but there's still a few little things that it kind of skipped over that I wish it did. They're both really good though. I feel like 5.6 six, especially Soul and Luna models are going to be like my main go-to. But then when I really want like a second opinion or wanted to think through some really really tough problems, I feel like I would go to Fable more for that. But like 5.6 will be my daily driver and Fable will be like my extra thinking power that I need. Now saying that, I did build most of the new version of the Future Tools website with Fable. And then I gave everything that Fable built to GPT 5.6 6 and GPT 5.6 found a couple of security vulnerabilities that Fable missed when building. Like it found some things that would have accidentally exposed my API key if people looked in the right places and some little stuff that would be like stupid stupid mistakes that Fable missed, but GPT5.6 found. So, I feel like they're going to be like really good in tandem. And I'm really excited. Like this is this is some of the most fun I've had in the last like I don't know year and a half. Like the models are so much fun to just build stuff with right now and I'm so excited to play with them and like this is the worst it's ever going to be. This is just crazy. I'm nerding out a little too much here.

Just a quick reminder, I do record these on Thursdays, publish them on Fridays. So if any new news came out like late Thursday night or on Friday, it'll probably make next week's video. But my goal with this channel is to drink from the fire hose all week, read all of the news that comes in, test all of the tools that are there, play with the new image models, play with the new video models. Like, there is so much coming out. And I spend my week just using and testing and figuring out what it's good at and finding out what other people are saying about them so that I can turn around and make one video for you at the end of the week that tells you just what you need to know. You probably don't realize this, but I record way more news stories that don't make the cut. And then when I watch a video back, I go, "No, that's fluff. No, that's hype. No, that's not interesting. And I just cut down to the stuff that I think the most people want to know about. So, if that's something that's interesting for you, maybe consider liking this video and subscribing to this channel. I'm getting really close to a million. It would help me out a lot. But yeah, that's what I got for you today. I'm also starting to do more and more tutorial videos midweek where I actually show practical use cases. So, those have been really fun as well. So, uh, I'll put one in the end card here somewhere, so you can see the most recent one that I did about using AI to create visual effects for videos. It was really fun video as well that I think you'll probably enjoy. But that's what I got. Thanks for nerding out. Thanks for hanging out. Hopefully, I will see you in the next one. And yeah, really appreciate you. Bye-bye. What's everyone doing? WELCOME TO THE SHOW.