MYTHOS is LIVE!!!!

Oh, it's not >> all right. Welcome, welcome, welcome, welcome. Yes, Fable is here, aka Mythos. I'm just pulling up some notes. Give me a sec. What's up, Greg? Marvin. Let's see if we can get Cloud Code uh working. Yep. Fable 5. Got it. All right. My goodness. Okay. So, I've had access to it for a week. I've been playing with it. I have a lot of thoughts about it. I'm going to be recording a video during this stream. We're gonna be having fun. What's up, Evan Steen? Good to see you again. Uh, before we get started, like the stream. I would really appreciate it. Thank you in advance. What's up, Nicola from Greece? Yeah, buddy. Slovenia. Nico, Slovenia, late to the M. Michael Chancellor, you are not late. We just started. Welcome, Steu Bonum. It's no longer too good to release to the public. That is true. Crazy how things change, huh? All right, let me load up my OBS and we shall start recording in a moment and I'm going to tell you all of my thoughts from using it for the past week. I'm I'm not even I'm not even going to hide it. It's incredible. Um, it it has some quirks certainly and I'm going to talk about it all. Okay, click. One, two, three, four, five. All right, we are now recording. How's that looking? Looking good. All right, so yeah, Mythos is definitely here. Um, let's see. I have too many windows open. Mythos is here. Mythos is here. That's Yeah, start building quick, please. Actually, that's a good call. I'm going to kick off some things uh just so we can test it while we're talking. Um so, let me do that. I don't want to burn through all of my tokens, though. So, I have to be a little wary about that. I'm going to kick off four tests that we came up with yesterday for it. Um, oh, interesting. Wait, no, that's Opus 48. I just saw Mythos. Where did it go? Am I missing it? It might have disappeared from cloud code. I swear I just saw it. Okay, I'm going to reopen it. Fable. Fable is a good model. Yes, it is a good model. So interesting. It is no longer in the model dropdown in cloud code. Maybe in Claude Co-work. Yes. Interesting. Uh you guys can't see this, but it's showing up in cloud co-work. It is not showing up in cloud code, surprisingly. So, we might have to save some of these tests for later. Uh but as you can see here, we have fable fable high. We can set different effort levels all the way up to max. And I'm going to explain all of it. Well, I see the system card, but we don't want the system card quite yet. Fable anthropic blog post. Let's see. There we are. This is what I want to see. All right, let's do dark mode. Dark mode everything. Okay, we got the blog post going. We have everything. We're going to get started. Uh before we get started, thanks all for joining. This is going to be fun. We've been waiting for this for a while. We've been talking about Mythos. We've been talking about what eventually would come out. So, I'm excited it's finally here. I'm excited to show and share what I've been experiencing over the last week. Okay. So, Mythos is finally here and it has come in the form of a new model called Fable. Fable is I guess you can say kind of the child or the sibling of mythos. It is effectively mythos but with additional guard rails. So mythos is given to the security community. Fable is given to everybody else. And the short and the short of it is it's an incredible model. It feels very different than any other model I've tested. It feels bigger. It feels like it can take on kind of more complex tasks, run for longer, longer time horizons, but it definitely has a bunch of quirks to it. And I'm going to go over all of that. So, here's the tweet right away. Introducing Claude Fable 5. This is Fable, the new family of models. So, remember you have Haiku, Sonnet, uh Opus, now we have Fable. that is the the brand new generational model trained using or or trained in the class of mythos model. So a mythosclass model that we've made safe for general use. Its capabilities exceed those of any model we've ever made generally available. Oh, Alex Albert, very cool. Uh okay, let's look at the benchmarks. Everybody wants to know about the benchmarks. That's what we're going to be talking about first. and and uh yep, I got that. And uh and if you're watching this video and you like me breaking down all these new model releases, please like and subscribe. It very much does help. Thank you in advance. It is quite hot in my room. Okay. Can you all see the benchmarks? Let's see. Resize the fit. Yep. Okay, that looks good. All right. Okay. So, here we go. We have Claude Mythos 5 and Fable 5. And they've basically grouped them into one category. Remember, the only difference between Mythos and Fable is that Fable has guard rails. Mythos has those guard rails removed. Mythos is given to the security community to help harden software, to help find bugs, to help find zero day vulnerabilities. And Fable on the other hand is kind of with guardrails won't do those types of things, but it's good for everything else. And what we see is the benchmarks show a lot. Now I know all of us think about benchmark. Um let me try to restate that. Uh one second. Okay. Um where was I? Right. So when we talk about benchmarks, I know benchmarks can be misleading a lot in a lot of ways. I know we look at benchmarks and it seems like every model release just increase the benchmark, increase the number, number goes up no matter what. But even when we use it, the vibes of the model don't reflect what we're seeing in the benchmark. And that's definitely the case. Maybe with the exception of deep, which is a benchmark that came out a few weeks ago. Unfortunately, I do not see that benchmark here. Uh, so we're just going to have to go on kind of the more traditional benchmarks. Aenta coding, SWEBench Pro 80% as compared to Opus 4.8 at 69%. GBT 5.5 at 58%. Now, here's the interesting thing. Everybody that I talked to says GBT 5.5 is more capable than Opus 4.8. However, on Swebench Pro, we're seeing a 10-point differential. However, and we're also seeing another point uh and we're also seeing another 10-point differential between Claude Opus 48 and Mythos Fable. Um Agentic Coding Frontier Code Diamond 29 uh 29.3%. Look at this. Opus 4.8 half of that. half of that GPT 5.5 5.7 it doesn't feel like these benchmarks are accurate to what I was feeling using the actual models between Opus 48 and GBT 5.5 but again let's let's just keep going we'll we'll hopefully deep suite comes out soon so GDP val this is a benchmark created by open AI GBT 5.5 1760 69 versus 1890 with Opus 48 now 1932 and GB GDP valve tests real world knowledge work knowledge work vision nice bump spatial reasoning interestingly for spatial reasoning GBT 5.5 did really well claude opus underperformed but now we have a new frontier model 38.6 six with fable. Tool use got a nice few percentage point bump here. Computer use 85% as compared to 78 and 83. To be honest, GPT 5.5 feels like within the context of the codeex app. The best computer use model, the best browser use model that I've used. legal agent benchmark 2% versus 10% versus now 13%. Humanity's last exam, the the most scarily named uh benchmark on the planet. We now have two new Frontier models, two new gold first place models, uh with a slight with a slight decrease with tools from the preview to what we're seeing today. Here's terminal bench. A very very important benchmark if you're doing a any type of agentic coding. We have 83.4 versus 88%. Okay. Very very good. Um it is a fantastic model. Okay, I'm going to share my actual thoughts of using it over the last week. Just give me a few moments. I want to get through some of this initial information. All right, let's look at the blog post. of course, you know, take everything they say in the blog post with a grain of salt. Um, obviously they're going to talk it up. Obviously, they're going to say it's the best model in the world. I can tell you from firsthand experience, it is a phenomenal model. It is more like it is more different than any other model that I've ever tested. A lot of the times it's better. Sometimes it's just weird and has these quirks to it, but it is it feels very different. It feels like a new training run. Okay. And remember, this is a 10 trillion parameter model, the first one of its kind. All right. Fable 5's capabilities exceed those of any model we've ever made generally available. It is state-of-the-art on nearly all tested benchmarks of AI capability. Now, I cannot wait till it gets tested on Deep Suite, showing exceptional performance in software engineering, knowledge work, vision, scientific research, and many other areas. The longer and more complex the task, the larger Fable Fives lead over our other model over our other models. Now, I want to pause here. This is what I saw most of all. No matter what task I gave it, it felt like I was kicking off a massive exploration. Fable wanted to look at my entire codebase, consider every single possible angle of every line of code that I ever wrote. Maybe it even checked projects that I haven't touched in years. It really just felt like I was kicking off this entire exploration. It was really It's such a weird feeling because you give it a small task and it no longer felt small once you hit enter. It really had this eagerness to check every single angle on any uh on any task that you give it and it was very capable. Now when I did give it tasks that I thought were extremely complex that would require long time horizons to complete, it was it was like no problem. It it didn't hiccup. it didn't it just it was like okay I got it and in fact even the most complex task that I gave it it it was almost like uh insulted by how simple it was. That's really what it felt like. Um and and again this is like some of the weird quirks with its personality which I'll explain in a moment. I could not find a task too complex for it. Let's just say that there was no task that I could come up with where it stumbled. Uh so to release the model both safely and quickly, we've tuned these safeguards conservatively. They'll sometimes catch harmless requests, though they trigger on average in less than 5% of sessions. Now, during my testing, I did not experience this once. I did get a heads up that the model might have many more false positives than what I was used to. I didn't have that once. Now, I, you know, I wasn't really explicitly telling it to look uh over the security of my application, try to come up with ways to improve it, but I didn't experience that once. Hey, Senator, thank you for the 10. Fable Mythos might be an information sponge. Be careful not to become the product. It's it I don't think that has to do with Fable. I think that's any model. If you're if you're chatting with it, you are the product. You're giving it your data. It's going to be very interesting to see how how this progresses. Um, okay. Thank you again. All right, let's keep going. H and then we're going to do some testing. Hopefully, it starts showing up in Claude Code. Do you see it yet? No. All right, let me check one more time. I don't know why it's showing up in Claude Co-work. It's showing up in Claude, but not Claude code. Very interesting. Very interesting. No, not at all. So weird. All right, let's keep going. Let's talk about the pricing. If you are surprised by the pricing, don't be. We thought I actually thought it was going to be more expensive. Um, it is coming in. So, Fable 5 and Mythos 5 are being offered at $10 per million input tokens and $50 per million output tokens. This is incredibly expensive. But here's the thing, you don't actually need Fable. You don't actually need Claude for the vast majority of use cases. If you've been watching the channel, you know, we've been talking about model routing a lot. You know, we've been talking about efficiency, the kind of multimodel world that we're almost definitely going to be experiencing in the coming years. This is a perfect example. You have the absolute frontier with Fable and you give it your most difficult problems and you pay that $50 per million output tokens and you say, "Thank you, sir." And then for everything else, you don't need it. You can go back to Sonnet. You can go back to Haiku. I really encourage you to think about what task you're assigning to which model. And and the more you know about that, the more prepared you're going to be for the coming years. We're we're already seeing companies b at these crazy bills that they're receiving from anthropic and open AI. So if you know how to route the models properly or sorry, if you know how to route the task, uh if you know how to route the task properly, you're going to be in a good position. Uh, and it is less than half the price of Claude Mythos preview. Very interesting. Um, but I will say it feels very slow and so I'm going to go over everything. Ouch. On pricing better be worth uh Yeah. Yeah. I I feel like $200 a month is not going to be enough. All right. So, FA uh Fable 5 and Mythos 5 can work autonomously for longer than any previous Claude models. Below, we discuss how these skills apply to software engineering and cover the model's improved capabilities in knowledge work, vision, memory, and life sciences research. This is what I noticed most of all. The model was so capable of doing tasks for long periods of time. And as I mentioned, there really wasn't a task that I gave it that it turned around and just gave me a quick answer or turned around in in like two minutes and and gave me something. It never did that. It was, you know, five plus minutes minimum. Uh during early testing, Stripe reported that Fable 5 compressed months of engineering into days in a 50 millionline Ruby codebase. Shout out Ruby. The model performed a codebasewide migration in a day that would otherwise have taken a whole team over two months by hand. And this is really it, right? Like if this is the type of task you're looking to complete, $50 per million output tokens is a bargain because otherwise you're going to be paying a ton of engineers stripe salaries to work on it for two months. And now that same team can oversee its progress and work on other things in parallel. Here's something else that I'm going to talk about. Fable 5 is also more token efficient than past cloud models. This is very interesting. Now, let me share my experience. The information density coming out of Fable was unlike anything I've ever seen. And so it might be more token efficient from an algorithmic standpoint, but just simply from the output explanations of what it was accomplishing for me, the information density was it hard to read at times if I'm being honest. Like it was using extremely complex words, extremely complex descriptions. It was very verbose, but not only verbose, it was information dense. And so when you have information density, basically more information getting conveyed in fewer words, fewer words, it is effectively increasing the intelligence of the model because then with those fewer words you can do more. Thus you can throw more inference at it or sorry you can throw more compute at the inference. it can run for the same amount of time and get more out of that same compute time than it would for Opus 4.8. So that is really interesting and it actually leads me to think about something else. This is kind of I think way in the future and maybe we won't allow it at all. But information density seems like something we're not really talking about all that much. How much information can you get out of every word conveyed by the model? Now imagine there was a way to increase information density even further. Now for me reading the output of Fable, I actually found it very difficult. I had to slow down my reading pace. I had to really think about every word that it was telling me. But another model doesn't have to do that. And so there's this argument that fable or future AI models might actually develop their own language, a hyperinformation dense language that only it can read. Now obviously there are some big problems with that, some big risks. If we can't read it, then we have no idea what they're talking about. We have no idea what they're planning, what their intentions are. And so that becomes extremely risky. But there is this argument that we would actually have a much higher efficiency and be able to get so much more out of the models if they simply communicated in maybe a non kind of alpha numeric language symbols maybe I I don't know it's so interesting to think about though and I only started thinking about this because of the information density out of the models. So on Cognition's Frontier Code Evaluation, which tests uh which tests whether models can pass difficult coding tasks while meeting the standards of high-quality production code bases, Fable 5 scores highest among Frontier models, even at medium effort. Okay. Um I want to talk about the effort. Now, I tried two different effort levels with this model. Um, let me look up what they were just so I make sure I name them properly. Okay, so we have I I basically tried two different effort levels and really it's just one effort level but with some extra sauce on top. So I had most of all I used the extra effort level and it was just overkill. It was slow and it it felt like it was way too powerful. It felt like I can dial down the effort on the model to the lowest possible thing and it might even still be too high of effort for what I needed. That's it's crazy to think about. Then Ultra Code. So, Ultra Code, if you're not familiar, is their new workflows feature. It basically kicks off a planning agent which delegates out to potentially hundreds of sub agents in parallel. And I saw this live and it is crazy to watch live. You see uh you see it start to think maybe a few minutes of just planning and then it would literally I I gave it a task of um review my entire codebase, give me a report and it would almost spin up an individual sub agent for every single file I had in my entire codebase and I would see all 100 plus of them running in parallel and just obviously watching my token budget just explode. But fascinating to watch and it feels like Fable is really good at utilizing the workflows feature parallel delegation of agents. Um it's not called ultra think it is called ultra code. I'm looking at it right now in cloud code. Uh they said it should be showing up in cloud code soon because cloud code it is not yet available. Uh fable. Yep. Not yet. Not yet. Okay. And so I encourage you as you're using this model, as you're using Fable, start on the lowest possible thinking effort setting and most likely it's going to be sufficient for your use cases. Here's Frontier Code. Uh mean cost per task, log scale. This is the score. Here we go. Yeah. So, one thing is interesting. I think it was uh who was it yesterday who had that great blog post? Uh Gnome Brown from OpenAI maybe. Polomial. I think it was Nome Brown. Yeah. Polomial. I think this is who posted it yesterday. Right. So, Gnome Brown who is let's see. Yeah. Nome Brown who's a researcher at OpenAI talked about this yesterday. He wrote this fantastic blog post about there really doesn't seem to be a limit in thinking tokens to quality uh relationship. Meaning you can continue to throw tokens at a problem and it will continue to improve the output. There doesn't really seem to be a limit or at least we have not found it yet. And so what you're seeing here, let's see if I can make sure you can see this on the screen. Yep. Okay. And so what you're seeing here is even when you kind of come up on the 100 million tokens, the models are still improving. the models are still improving which means like there they're there they're there just doesn't I mean it just continue to throw tokens at these problems and it seems like the compute demand might actually be higher than we even anticipated which was already high to begin with. So something to keep in mind. Cloud code available cloud update. Okay. Uh Dorian, let me try to update it real quick because I do not have it. Uh check for updates. Yeah, no, I'm on the latest version. It's just not appearing for me. It appears in the desktop app for um cloud kind of chat and co-work, but I'm not seeing it in cloud code. Let's see. No, nothing yet. All right. So, this was a great blog post, by the way. I'm going to drop it in chat if you all want to read it. Cool. Uh, all right. So, let's keep going. fantastic blog post and I think Mythos Fable really emphasizes this point where there doesn't seem to be a limit in the number of tokens you can throw at a problem. Um, and Fable in particular seems incredibly willing and eager to use all of those tokens. So really like dial down the effort and only dial it up when you need it for sure. Um, thank you Nizzless. Thank you, Nizzilless. Thank you very much. Okay. Uh, let us keep going. We got the system card. Okay. Yeah. So, what we're seeing is like uh the the tasks continue to get harder. the cost of completing those difficult tasks can continue to go up, but there doesn't seem to be really a hard limit anytime soon, which is crazy because test time compute will just continue to scale. Pre-training will just continue to scale. We have all of these new scaling laws that are building off of each other and the models just kept keep getting better. Um, Seth. Hey, Seth. Uh, slashmodel. I'm not in cloud code uh CLI. I'm in the desktop app. I want to see it in the desktop app. I am a UI guy. Okay. Uh, here's a video. Claude Fable 5 beats Pokemon Fire Red only using vision. Wow. Full speed on this one. No maps, no navigation aids or extra game state information. Earlier Claude models needed a complex helper harness to play Pokemon. Cloud Fable 5 completed the game with vision alone. Very, very cool. Here is a simulation of the solar system and predicting solar eclipses. This doesn't seem all that crazy. We've seen stuff like this. Here's the problem with showing off demos of the model. All the models today, Opus 4.8, GPT 5.5, they are very capable of building exactly this. That's actually one of the reasons I stopped doing model tests in my videos because I couldn't even come up with tests that were difficult enough for it, which is kind of wild to think about. Like there was no test that I could come up with where I was like, "Wow, this is a standout model because it was able to achieve this thing." The last time that happened and really one of the last model tests I ever ran was for a Gemini Pro uh Gemini 2.5 Pro where it was able to simulate a Rubik's Cube and actually scramble it and solve it correctly. That was really the last time where I was blown away by something a model was able to do that previous versions were not. Now it's all about the vibe of the model. It's you have to actually get in and really start using it and see where those kind of edges are. And I haven't found any with Fable 5 yet. Um, okay. Let's see. Uh, Mythos 5, our internal protein design experts accelerated aspects of the drug design process by around 10 times. Very interesting. One other thing I want to talk about that's super interesting is their release and fear-based marketing. And so we got reports, right? So, it was like a few months ago that we heard about Mythos for the first time. And then we just got reports that Anthropic has been testing Mythos since January. And even though we knew about it publicly when it Mythos was announced, nobody had it. And it wasn't until now June 9th, 2026 that we actually had access to the model. I think this was very intentional. I think Anthropic wanted to keep the model to build their next model. That is the reason they want to accelerate their own development. They want to accelerate their own research to the point, excuse me, to the point where they feel comfortable that they have a sufficient lead and then they can release the model to everybody else. And they have a history of this. Uh if you remember, they cut off XAI's access to the cloud family of models because they didn't want their competitors building using their own models, using their own technology. So they have a long history of this. So something to something to keep in mind. Uh okay. Sure. Barbecue slashmodel cloud fable 5 include desktop app code works. Um maybe let's let's test it out. slashmodel um Claude Fable 5. Yes, fantastic. Sure. Barbecue, thank you. Thank you for that tip. We do now have it. Let me just show you in Claudes. Actually, I'll come back to that. Okay, while we do this, thank you for that. Um, Fable 5, but I I don't know. Is this Can this be real? I can't select I can't select fast mode. I can't select the thinking budget or the thinking effort I should say and it then it just disappears afterwards. Interesting. Slashmodel um see what it was. Claude Fable 5. Claude Fable 5. Yeah. Okay. So, it did work. Uh, I'm going to kick off a few. Let's see if it actually is doing it. Yeah, it seems to be running Cloud Fable 5. Okay. Slashmodel Claude Fable 5. Yeah, so that is working. I can't set the reasoning. Uh, but that's okay. I'm going to kick off a few tests. Hopefully, it's actually using it. I know you guys can't see it, by the way. Um, so just bear with me. I'm going to show you this all in a few minutes. I'm going to kick off five four different tests right now. Uh, model Claude Fable 5. And it's already doing the thing where it's asking me a bunch of questions, which I'm going to come to in a moment. Um, okay. So, I kicked off just build. Don't ask more questions. Just build, don't ask more questions. Okay, so I kicked off a few of these. Uh, I think this is all kind of interesting. New safeguards, safety classifiers. Um, what is this? Okay. Interesting. Listen to this. So, when they're talking about the safety of the model, we've previously identified large-scale attempts and they link the article in which they talk about it and we covered it in a video on this channel. Uh, yes, we're going to test it everybody. I promise we will test it together. Uh, drop your any prompts that you want to see tested, drop them here. The only problem is tests take a long time, okay? Especially for Fable, it takes a very long time. So, drop them now. I'll select a couple and I'll throw them in. But let's keep going. Um, to extract Claude's capabilities to train competing models in authoritarian countries, distillation of Fable 5's abilities could indirectly lead to the proliferation of nearfrontier AI capabilities. We already have near Frontier. Now that I've seen Fable, I think open source is more than 6 months behind. I think it's probably closer to a year behind. and these could be released without the appropriate safeguards. Requests that are flagged by our classifiers as being part of a distillation uh as being part of such distillation attempts will fall back to opus 4.8. Whoa. They're like, "Oh, you want to distill? Go ahead. You can have our old model." Opus 4.8. That's so funny. Prompt solve the Iran conflict for sure. Yeah. No, it'll do that. That's That's too easy, I think. Uh, a new data retention policy. Let me say that again. A new data retention policy. Uh, changing the way that we handle business customer data for Fable 5, Mythos 5 and future models with similar or higher capability levels. We will require 30-day retention for all traffic on Mythos class models on both first and third party services. We won't use this data to train new cloud models or for any non-safety related purpose. And we've instituted a new we've instituted new privacy protections including logging all human access to the data and ensuring its deletion after 30 days in almost all cases. And why do they do that? No, they are not using it to train. Although if you're a direct customer, not a business customer, maybe they are. But this data will help us defend against complex and novel attacks including new jailbreaks. Uh Ply, I'm sure you're going to love this. and attacks that operate across many requests as well as help us identify and reduce false positives. Very cool. Um, okay, Claude, let's see. Fable 5 is available everywhere today. So, if you want to try it, you can try it right now. Pricing for both models, $10 per million input, 50 per million output. developers can use Fable 5 in the cloud API. Um, we expect demand for Fable 5 to be very high. Now, here's the thing. As I was using it, it did feel very slow, and I wonder if that's a function of just the size of the model, also how many tokens it kind of was eager to use, how thorough it wanted to be. These are all part of what makes it feel slow. Now, I want to go over I wrote a a little review on Twitter. Uh, by the way, um, uh, if you don't mind, please like the stream, subscribe to the channel. I would very much appreciate it. And if you're watching this, uh, on demand, please like the video. It very much does help get the word out there. Thank you in advance. Um, so I talked about the model being really good. Let's just get that out of the way. It is incredibly good. It is incredibly good. Um, it found I like my one of my favorite prompts to give a new model that I'm testing is review the entire codebase and it reviews it for security. It reviews it for documentation, logical gaps, edge cases, UX, UI, workflows, test coverage, everything. And it found things that no other model saw. And I will say this, it actually said that that my code was in a really good place already. So, it found things. It didn't found it. It didn't find a ton of things, but it did find things. And all of it has been built with either Opus 48 or GPT 5.5. So keep that in mind. But I know all of you know that the model is good or at least like I've said it enough times. Let's talk about the quirks. As Doug Demiro say as Doug Demiro says the quirks and the features. It is incredibly verbose. I already talked about this. Explanations get super deep, super technical very quickly and I had to update my Claude MD file multiple times to try to encourage it to simplify its explanation to me. I would often, more often than not, have to tell it, please simplify the explanation. I felt dumb reading uh Fable telling me what it just did. I felt dumb. It was not a good feeling. Right? This is the first time where I was like, "God, can I just explain it like I'm five and I I've never really had to say that to another model." And I talked about the information density already. Uh I talked about the fact that there's an argument because of how well information density works to increase efficiency of a model. Uh maybe agents will develop AI will develop their own hyperdense language. Um here's another really odd thing. It wanted to ask clarifying questions so very badly. It was so annoying. And in fact, I spun up Cloud Code right now and it's asking me uh just build it. Don't ask more questions. It is asking me questions right now for those four prompts that I just kicked off. It's asking me questions and and like yeah, I'm okay with a couple questions for complex tasks, but no matter how difficult the task, it was asking me to clarify questions. So, here was the flow. A single prompt would turn into it would ask me, let's say, three to five clarifying questions, sometimes more. Then it would summarize my answers and ask me to confirm the summary. Here's what you said. Is that what you just said? Okay, great. Then it would say, I'm going to write a speck. Is that okay with you? Yes, just write the spec. Okay, here's the speck. Is the speck correct? Please read it. Yes. Yes, the spec is correct. Keep going. Go. Okay. Then it would ask me to confirm its agentic approach. Should I spin off a bunch of agents in parallel or should I should or should I do it sequentially? I don't know. You decide. Why are you asking me these things? Just go build. And I got so frustrated. And then finally, after all of those questions and confirmations, finally, it would go build it for me. And when it finally did, it was great. But that was very frustrating to watch. And then yeah, finally uh it just felt very slow. I've already talked about this. It was slow to start. I would watch. So So in Cloud Code Desktop, there's two things that it shows you really. It gives you a timer on how long the task is taking and then it tells you how many tokens it has used. And often what would happen, especially within the first five minutes, it would go to like 1,500 tokens. That's 1500500. and the timer would just tick up and I'd be like, "Is it doing anything?" I couldn't figure it out. I don't know what it was actually accomplishing during those first like five to eight minutes. And this would happen on every single prompt. I would just sit there kind of scratching my chin thinking, I is is it going? Like I I wanted to poke it. Uh, and then like in a blink of an eye, especially in workflows mode, it would go from like 1500 tokens to 1.5 million like like within 30 seconds. It was it was crazy to watch. Okay. Okay. Who wants to see it tested? I think I've talked enough. Um, one one last little thing. I promise I'm going to get to the testing after this actually cuz I have I have some notes that I'm looking at over here. Uh, I want to talk about loops. I just made a video on loops. If you don't know what a loop is, it is the thing that Peter Steinberger, it is the thing that Boris Churnney has been talking about. The next abstraction layer above agentic engineering. Uh, I just made a video. If you haven't seen it, please watch it. Um, it's it's much simpler once you understand the concept, but when I think about loops, workflows, and fable together, it really feels like nobody understands what's coming. I can't stress that enough. When you have the ability for the model which is already incredibly autonomous and then you also parallelize it with workflows and then you wrap it in a loop in which it will just continue to burn tokens trying to build whatever it is you set it to build until it reaches some goal. I I just can't imagine how powerful that is. the idea of building software factories is very much here and only a fraction of everybody doing uh agentic engineering is even scratching the surface of what's possible. I I think not even anthropic and open AI are fully utilizing what is there. the model overhang, the concept of the models being so good and we don't even know how to get the most out of them is so very real and it became much more visceral to me after using Fable. Now with that, yes everybody, I will do tests. Let's test. Let us test. Let me throw up Claude. Okay. I mean, this is a perfect example. Like it just says thinking I don't know how many tokens it using. This is this is actually different from the interface that I had before. Uh let me just tell it to go. No API key on this machine. Add your key. Okay. All right. So, it's still doing the city builder game. It's still doing Oh, interesting. Interactive. I'm seeing so many bugs here. All right, I'm going to end my recording. Start uploading that. Just give me one second. Just some behind the scenes stuff right now. Click. One, two, three, four, five. Okay, let us test. I'm going to restart my camera because it is laggy. All right. Slightly better. Not really. Got a hair sticking up right here. All right. So, let's um let's keep going. Um here we go. So, interactive 3D Rubik's cube. I don't know why it's pulling up this. Let's try to do it. Well, that's a bad start. I don't see anything here. Can you all see this? I I can't. There's no Rubik's cube. That is so very disappointing. Um, okay. There is no Oh, and it switched to 4.8 again. Maybe this is 4.8 and it didn't actually update. Um, I am going to restart Claude uh desktop. Let's see if that worked. Nope. I am still not seeing it here. Look at this. I'm still not seeing it in this window right here. I wonder why. Um, I am going to I'm going to ping the team there and see if what's going on here. Does anybody have it in cloud code? Is anybody actually using it there? Uh, Greg Barry, thank you for the super chat. everyone else in the desktop app. You can type uh slashmodelclaudefable 10 and it will show you the model is fable 10. It's not fable. Yeah, I don't actually think the tests that I just ran were actually the model. In fact, great point, Greg, and thank you. H yeah, there it is. So it's not actually the model. You can name the model anything. It falls back to 4.8. Yeah. So I just called it the smartest model ever and it's not there. So I can't even test it. We do have it in cloud co-work though. That is for sure. There it is. And interestingly, look at this. Included until June 22nd. So what should we build with Claude Co-work? Anybody have any prompts they want you want me to test? I don't even know what to test in uh in here. Slashmodel Batman. Yeah. Um write a two sentence horror story. Launch it in CLI. Uh let's see. Maybe maybe CLI. I'm so bad at the CLI though. So, um let's see. Claude, let's see if it comes up at least. Yes, I trust this folder. Uh, Fable 5 is now available. Run Claude update. Okay, let's update Claude. I might have to use Claude code. Um, which is definitely not my preference, but here we are. Guess I trust this folder. Meet Fable 5. So, uh, model fable 5. There we go. Okay. I guess we're using cloud code. Let uh cloud code. Let me switch it. I turn. Here we go. We got it. Oops. Um, can y'all see that? Okay, we're we're we're testing it one way or the other. Please bear with me. Uh, I am not super familiar with Cloud Code CLI or at least I'm not proficient with it. Um, what what should we test? I'm going to grab some of the previous tests that we had planned. Uh, and we will test. So, here's the first one. Build an interactive playable 3D Rubik's cube that runs in the browser. I should be able to scramble it, turn the faces, and solve it. Make reasonable choices on the rest. Don't ask me any questions. All right, let's see it. This is our first live demo of Fable. Okay, so we see the tokens ticking up. thinking with high effort. Um, the loop feature, they're really pushing that hard right now. Set effort. Yeah, I'm probably going to set the effort lower on the next one. Uh, I think I have to wait till this is done, right? X high fire says sains cobar. Hi all, what's up? Mark Santos, what's up? effort ultra code. Who wants to see ultra code? All right. So, this is what I'm talking about. This is exactly what I'm talking about. So, we see the tokens creeping up kind of slowly, relatively slowly. It's only a minute in, but this is what I would see in the Cloud Code desktop using Fable during the testing period for the last week. I would see the uh the token stop at a few thousand and it would just tick up on the timer. It would get up to a few minutes all the way up to like 8 minutes before it actually started exploding with usage. So, we're kind of seeing that here. Almost done thinking. Okay. So, I guess it's just thinking. It's planning at this point. Trying to predict the FIFA winner, Musical Paradise. Yeah, maybe inference batching bottlenecks. That's what's stalling the token count. Yeah, possibly. But like this is quite slow. 7,000 tokens after two minutes. What's that in tokens per second? Uh let's say yeah, I mean it's just super slow. Um I'll kick off another one in parallel. Okay. So, we'll do a few of these in parallel. And it's interesting because we actually get to see what it did with Opus 48 because we kicked them off in uh Cloud Code Desktop. Um so, slasheffort. What should we do? Ultra code. Should we use ultra code? Do I want to burn through all of my tokens? Um, who who says we should use Ultra Code for this prompt? Let me show you. Build an interactive real-time fluid simulation that runs in the browser. I should be able to push the fluid around with my mouse and watch it react. Choose the method and controls. Add lots of settings dials to it. Don't ask me any questions. Okay. Yes, everybody wants to see Ultra Code. Okay. All right. We're doing Ultra Code. Rip my budget. Uh, here we go. We're on Ultra Code. There it goes. Let's watch it work. Okay. Seasoning. We see the initial token spin up pretty quickly. It should be able to do it without ultra code. Totally. Yeah. But what's the fun in that? We want to see it spin up 100 agents in parallel. And the way that I've gotten it to actually spin up the maximum amount of agents, sub agents, is by saying do a full codebase review. And it does it. Oh, I forgot to write make no mistakes. Oh, no. All right. Well, hopefully it doesn't. Okay, so it's still going. Three and a half thousand tokens so far. Uh, run clawed agents to see them all in one place. Okay. Interesting. Claude agents. Did I not do that right? See them all in one place. Didn't it just say do cloud agents clad? Do I do it here? Nope. That's not right. I'll build this now. Now a GPUbased Navier Stokes fluid simulation in a single HTML file. Is it still running? Did I cancel it accidentally? Yeah, I did. Uh slashworkflows. No dynamic workflows in this session. I know I'm using this wrong, by the way. I know I I seem like I don't know what I'm doing here, and that is because I don't. Um, if I can just get Cloud Code Desktop to load with the actual model, that would be so nice. So, very nice. But still not there. Still not there. Darn. Okay, I guess we're just going to have to come back to that. Yeah. And if for those of you just joining now, we are testing. We are doing some live testing. I feel like I'm not doing this properly. Oh man. Oh yeah, it's going to ask me to do all of this now. Oh my gosh. All right, I'm going to have to babysit this. Except it edits on. How do I uh shift tab to cycle? There we go. Okay. Shift tab in plan mode. Auto mode. Yeah, buddy. Auto mode. Um, shift tab. No. Do you want to proceed? Yes. Plan mode. Auto mode. There we go. I got it. How do I Why didn't cloud agents work? That's weird. Maybe I need to kick it off from claude agents. Is that right? Why Fable? Why not Mythos? So, Mythos and Fable are the same model. The only difference is Fable has uh stricter guard rails on what you can ask it. That's the difference. Good question. Yeah. Ancient Enigmas. Yeah, I use the dangerously skip permissions every time. Um, I I used to do that. I use auto mode now and it basically approves everything. I just I don't know. Let it figure it out. I don't think you're in auto mode. Okay. Pretty sure I am. So, reminder, this one should be workflows. But it says no dynamic workflows in the session. Weird. Weird. I don't know why this one's still working. Only 22,000 tokens so far after 5 minutes. It seems quite low. Marmar Labs, thank you for the $4.99. I know it isn't working in cloud code yet. They have an update uh to the app. They haven't pushed a cloud code yet. It's interesting because it is showing up in cloud code desktop or sorry cloud desktop. So the cloud and co-work but not cloud code desktop. Very very interesting. Thanks Marmar. Good to see you again. Um I really wish this was working. Okay. So, we're gonna do something else now. Um, let's do CD. No, not AstroHub. Let me double check where this is. Really? Oh, okay. Wait a second. We have a result. This literally just popped up on my screen. Here we go. Actually, let me make sure it was from this window. Yeah, here we go. Done. Okay. And it just popped up. Let me show you the results. The interactive Rubik's cube simulator. So, by the way, do you remember what it showed in Opus 48 that we just did it? Basically, the cube didn't even appear. What do y'all think is going to happen now? What do you think it is going to be? Um, give me thumbs up or thumbs down. You think it's going to work? You think it think it's good? I'm looking at it right now. Let's see what happens. Here we go. Bam. All right. So, let's see what we can do with it. Now, that is the most realistic looking Rubik's cube that I've seen. Okay, I can grab a side and I can rotate it. Very cool. Let's do scramble. Yeah, absolutely beautiful. So, the scramble worked. Now solve. Who thinks it's going to solve it? There it is. That is a stunning success. Um, super impressive. Yeah, I mean, look at the light, the reflection, the shadows. Really nice. Really, really nice. Um, isn't solve just reversing the uh ordering? It Yes, it could be that, but there is actually algorithms to solve it. Uh why uh write a two sentence horror story. Okay, I'll try that. Why? Why is that one so interesting? I'm going to put it on the lowest effort for this one. Write a two sentence horror story. Let's see how long this one takes. Oh, you're not looking at it. Uh give me a sec. I turn two. There we go. Okay. So, it is Okay, there it is. Yeah, it only took 4 seconds. Uh, the mirror in the hallway had always shown my reflection a half second late and I learned to live with it. Tonight, it moved first. All right. It's all right. Uh, now I'm going to try I'm going to I'm going to keep it on high for this next one. Switch to high. Uh, add a bunch of SL um of sliders, settings, and features. Okay, let's see if uh let's see what it comes up with there. Uh, use slashgoal ultra code fast fable 5 with 100 agents. Hack the Pentagon loop. Write a horror book. Let's see. Give me the other prompts. Make a 3D game engine. Um, you know what? Let's do a Minecraft clone. Oh, actually, okay. I want a Sim City clone. That's what I want. And I'm going to spin up a new version of Claude. This I really think we can get it to use a bunch of agents in workflow mode. Workflows. Okay. So, uh, let's say effort. We're going to turn it up to ultra code. Right. This is right. Build a Sim City style city builder that runs in the browser. I should be able to zone areas, place roads and infrastructure, watch the city grow and manage a budget. Pick the systems and features you think make it fun. Make it as visually beautiful as possible. Don't ask me any questions. Here we go. This is Ultra Code. It really should be using workflows now. Slash goalgo. Yes, everybody wants me to use slashgoal. I don't really feel like burning all my tokens in like in like 5 minutes or whatever it's going to take. Okay. Fluid dynamics. Should I show that one? Yeah. All right. So, My internet is working very slow right now. I'm gonna cancel this. I'm gonna uh give me one moment. I'm just going to re-upload. There's no reason it should be this slow. All right. So, look at this. It's still going. Almost done thinking with extra high effort. We did this one. This one was cool. Uh, bam. We'll do auto mode for this one. Fluid dynamic simulation. This one's still going. Uh adding more features to it. Yeah. Wow, it's so slow. Okay, we're waiting for this to finish and I'm just going to see if I can Wow, it is. Yes, my internet is super slow right now. Hopefully it's the stream looks okay. Or is it laggy? Oh yeah, it was Safari. Weird. Yeah. Okay. I tried to upload to Google Drive using Safari and it was going to take an hour and a half. I just switched over to Chrome and it's going to take one minute. I do not understand what that is. Safari maybe had some kind of like block on it. I don't know what happened there. That's so weird. All right. So, keep dropping your uh keep dropping your prompts. Keep dropping your prompts. We're going to do it. Uh one more time, if you can like the stream, I would very much appreciate it. Want a bunch of people in here. We have almost 3,000 watching right now, which is kind of crazy to think about. Thank you all for joining. This is so fun. Uh, dishwasher 69. Make a video using code. No theme, no direction, nothing. Okay, I like this one. Yes. Okay, got it. Yes. Yes. Yes. We're going to do that one. Uh, copy text. Make no mistakes. Make no mistakes. Of course. Of course. Make no mistakes. Um, trust this folder. We're going to do effort. I'm going to set the effort on this to medium. I'm just testing different effort levels. We'll see what happens. Uh, okay. I'm just going to clean up the prompt a little bit. Make a uh there's a bunch of characters in here. Make a video using code. No theme, no direction, nothing for me. Make it like it's the only thing you'll ever make. Put everything in it. I don't even know what that means. Let's see. Let's see what happens. Uh DJLA music, is it possible to access it today? Yes, it is very possible. You can go access it right now. You can use it with cloud uh cloud code. You can use it with cloud desktop, cloud co-work desktop, but not cloud code desktop. Um, let's find out what's going on. Okay, maybe cloud is now working. Let's see. Nope. Still not working for me. Yep. No. Uh, okay. So, Cloud Code Desktop is still not working. That is um so very disappointing. But we're going to be using Cloud Code CLI. Um, okay. Yes. And auto mode on. Don't ask me any more questions. Does anybody know what this caffeinate means up here? I've never seen that before. Caffeinate. What is caffeinate? Keeps your Mac on. But why? Yeah, I I don't have caffeinate. I have amphetamine on my computer. But why does it say caffeinate in this tab though? It won't sleep your machine. I see. Okay, cool. I didn't know that. Um, is it done? Interrupted. No, just continue. Okay, this is still going. Fluidbased simulation. I know Jonah just created one. Let's pull that one up. Um, where was that? Live streams channel. Whoa. Oh my god. Okay. Yeah. Yeah, this wins. All right. I'm going to show y'all. Hold on. What was it? Can you give me the prompt? Oh, you you did. Never mind. Okay. Okay. So, here's what Jonah just sent me. By the way, uh I he sent me this using here. Now, which is a sponsor of our channel. I just love them so much. Literally just say publish and it gives you a URL in seconds. Shout out to them. They're not sponsoring this. I just love them. Um here we go. My goodness. My goodness. Let me move my face out of the way. Look at this. I'm going to restart it so you can see. Look at this. Whoa. And you can see the light coming through. All right. So, the prompt for this, make the most amazing locally hosted fluid dynamic sandbox with ray tracing. Something that is amazingly satisfying to play with and incredibly visually appealing. Whoa. Very cool. Let's play with some of these uh effects. Let's turn down the swirl all the way. This is the best fluids dynamic simulator I have ever seen coming out of a model quality. Let's go to ultra. Awesome. Whoa. Oh. Oh. Uh, it just popped up a new Rubik's cube. All right, let's test this again. Uh settings, cube size, 5x5, turn speed, scramble moves, piece gap. What's that? Oh, interesting. Auto rotate. Oh, that's cool. Leave her auto rotate on. Field of view. We'll Okay. Sound volume. Color theme. Pastel. Nah. Neon. Let's do classic. Okay. scramble. There we go. And now I'm gonna solve it live. No, just kidding. Here we go. I'm gonna solve it. Yeah, really good. Really good. Um, not the best. Not the best I've seen. Uh, but in terms of the realism, in terms of it getting it right, it's an absolute uh absolute pass. Okay, back to here because this one's awesome. Uh, light follows cursor. Oh, look at that. So cool. Idle dreams. I don't know what that means. Uh, shadow depth. Bunch of different settings. Line height, specular bloom. Yeah, very cool. gold. Yeah, look at that. Inferno. Sure. Yeah. This is so impressive. I'm doing all of this. By the way, if you want to test it, I we literally have this live on the web right now. So cool, right? Yeah. I just dropped the link. If you want to test this exact thing, the thing that I'm looking at right now, you can do that. Oops. How did I get rid of chat? Uh h chat just disappeared. Give me one sec. Do you see this comments? Here we go. All right. All right. Yeah. So, go test it out. Shout out to here.now. Um, yeah. Very, very cool. I'm going to open. Let's test Claude one more time. See if we got it yet. Nope. Gosh, that's frustrating. Okay, we have some results back in Cloud Code. Let's take a look at that. Imagine coming into the stream right now and just seeing this. It probably make no sense if you're just coming in. We are testing uh the new mythos model uh fable. Okay, I got to get rid of this. It's going to be distracting. Um, okay. Where's the link? Uh, I will drop the link one more time if you want to see that fluid dynamic simulation. I uh the Rubik's Cube app 2. I don't know if I have here now installed into this, but yeah, this one's done. Okay, let's see. The full render is running in the background. It'll take several minutes, mostly the Mandelro Zoom frames. I'll assemble and verify the final file as soon as it finishes. So far, the pipeline uh is it done? 84 seconds 1280 by 720 30 frames per second. Let's see if it's done and then link me to or where is it? Where can I find the file? Let's see if I can find it. So, this is uh the original prompt if you're wondering. Make me a video. No theme, no direction, nothing from me. Make it like it's the last thing you'll ever make. Put everything in it. Literally no direction whatsoever. Uh, okay. I'm just going to take a quick peek at it before I show it on stream. Okay. All right. All right. You guys want to see this video? This is a video that was made with no direction. This is actually pretty cool. Um, by the way, I keep looking over there because Jonah is hanging out here. Um, let's see. Let's see. QuickTime player. Resize the fill canvas. All right. So, before I play this, let me just give the prompt again. Basically, make a video. No direction. Just go. Just go do it. No, I'm good. Okay, never mind. I'm not showing the video. Uh, okay. So, here we go. Oh, it has music. All right, hold on. Hold on. I got to pipe the music in now. It's the movie's still playing. And then it just finishes, I guess. Yeah, that's the end of it. All right, that is Did it freeze? I think it froze. Crazy. It froze quick time. Don't stare at it. Yeah. Oh my god. All right. That's That's a crazy make. I don't know. Oh, I don't know what to think about that. Wild. Just wild. All right, one more time. Let me see if it shows up in Claude. Maybe I have to uninstall it. That's so frustrating. I'm going to try uninstalling Claude and see if that'll work. Okay. Downloading cloud desktop. And nope, no fable. table. All right. All right. Um, what should we test now? What do you guys want to test? We have the Rubik's Cube. This is still going. 60,000 tokens so far. 25 minutes for 60,000 tokens. Seems quite low, right? That seems very low. Uh, it is still building my fluid dynamic simulator. It is. Oh, come on. All right. Auto mode is on. Create a comprehensive video. Yeah, we did this one. This should be Oh, yeah. This is Ultra Code. So, let's watch this go. Remember Ultra Code, lots of different sub aents in parallel. Uh, can you find some tweets or if anybody sees anything interesting? All right, let's uh go make create MacOSS goal create and make make no mistakes. Nice. Uh so let's see. No dynamic workflow. Why is the dynamic workflows not working? Why is it not spinning up multiple agents? It was so much easier to get it to uh work in Cloud Code Desktop. And apparently Cloud Code Desktop for Fable Fable within Cloud Code Desktop is still rolling out. That's why we haven't got it yet. P= MP. Yeah. Yeah. All right. That's the one. That's the one. Yes. I trust this folder. Uh, obviously obviously we have to change the effort. We're going to do max effort. Um, solve P equals MP. Make no mistakes. By the way, burning through my quotas for you guys. I'm going to see how much I've used so far. Oh, they changed the view of this page. Interesting. Uh, settings, usage. Yeah, it feels like an operating. Oh my goodness. All right. So, in my current session, I want you guys to guess how much of my current sess session quota have I used. I am I am on a max 20 plan 20x. Rosco, thank you for the $10. Solve colots conjecture. Um, okay, I'll spin that one up, too. I don't know how much quota I'm going to have left, but let's see. Tell me how much quota do you think I've used so far. Put it in chat. Let me know. We're going to do max effort for this one. I don't even know what this is. How would I even know? I don't even know if this is a real thing, but we're going to do it anyways. Max effort. Go. Okay. All right. So, how much quota have I used? How much quota have I used? There it is. 51% of my current session quota after everything we've done so far. Um 31% of all models, which resets on Friday, so it's only a few days away, but 31% is quite a bit for my entire week of usage. Uh, Spookxe, the Claude video one wasn't done. Possibly. Possibly that was it. Maybe it wasn't finished. I don't It's It's fine, though. Um, yeah. Look at that. And by the way, they changed the look of this, too. Uh, okay. Very, very cool. Um, let's see if any of these are done. This one's done. Okay, we got workflows. Okay, let me switch back so you can see the workflows working. I don't know what changed, but here we go. We have workflows. There it is. 63 63 different agents running in parallel. Look at that. My goodness. Look at all these tools. So, they ran each for, let's say, 2 minutes on average. By the look of it, they burned between 20 and 30,000 tokens each. Um, okay. So, the solve colots conjecture, we're 2 minutes in. Solve P equals MP. Darn it. Okay. Let's make sure we have auto mode on. And let's make sure we have auto mode on here. don't want it to ask any questions. You need to update cla code for uh no I am fully updated. It is not available yet. It is available in the desktop app for claude chat and claude co-work but not claude code. And uh I just pinged the anthropic team and they said that is expected. It is rolling out right now. Post the video. Dishwasher 69. Oh, we have Oh, okay. We have another fluid simulation. This is the one that just finished for my uh on my browser or sorry on my claude. Here we go. Oh, that's so sick. I think this is better than yours, Jonah. Look at that. Jesus. >> Moreh toggles, too. >> Yeah, more toggles. here. Sim resolution extreme. >> I did. >> Wow. Look at that. >> Wow. >> So cool. My god. Uh die resolution maximum. Yeah, man. This is by far the best fluid dynamic simulation I've ever created using a model. Wow. So cool. link. Uh, okay. Let's see if I can get this live and remember the fluid dynamic simulation was uh part of what used the dynamic workflows. Um, so let's say deploy this on here. Now let's see if it has access to here. Now I'm posting it. I'm post Oh, post the video, too. Uh, okay. I might post that later. Let's say time scale. Very cool. Velocity fade much. Okay, it fades much more quickly. Let's say boom. Di fade swirl vortex. Let's turn it up. Yeah, look at that. splat radius. Let's turn that up. Yeah, there we go. Solve capitalism. Uh spamming solve capitalism. Okay. Uh force fixed color mirror splats. Auto splats enabled. I don't know what autosplats mean. Whoa, crazy. Oh, blinding. All right, this is definitely a winner. Very nicely done. I'm going to turn off auto splats. Rendering dinormal pressure field. What? Oh, whoa. Okay. Velocity field. Dang. and vorticity field. Very cool. Bloom turned off. Bloom turned on. Oh yeah. Yeah. The Okay. The bit rate is uh dropping because uh there are so many pixels on the screen. Let's see if I can Let's see if this worked. Uh, okay. All I said was deploy this on here. Now, I don't have it installed. Oh, we got it. No way. That just worked. I It had no idea what here now is. Okay, not found. Hold on. Still going. Let's see if we can get it to work. And I'm going to switch back to term. I am now at 55% usage by the way. Uh yeah, that was that's very very cool. And during that time, let's see. Okay, so I'm going to see hopefully there. There we go. Oh, it just worked. Okay, I'm sharing the URL. There it is. It's in the chat. I'll drop it a couple times. There it is. If you guys want to play with that here.now shout out. There it is. Oh, you can't see that. Hold on. Hold on. Uh, back to Chrome. This is now live on the web. All I did was say deploy to here. Now that that's why it's one of my favorite one of my favorite products, just cuz I can just do this. Uh, deploy to here. Now, it figured out what here. Now is and deployed it. And so, if you want to play with this, I dropped the links in chat. Go play with it. This is awesome. This is so cool. Fluid simulation. Absolutely. All right, let me switch back to iTerm. Okay. Solve P= MP. Let's say I can't solve P= MP. And the only way to make no mistakes here is to tell you that plainly rather than hand you a proof that's wrong. I did verify before saying so as of June 2026, the problem is still open and the Clay Mathematics Institute $1 million Millennium Prize for it has sat unclaimed since 2000. Okay, so it could not do that. Solve Colex conjecture unsolved. I can't solve it either. At least it's not telling me it can end world hunger. Yeah, if only. All right. I want to say uh thank you to everybody who's joined. This has been awesome. I'm gonna put a video out all about Fable. It'll come out later today. Thank you for joining the stream. I want to just do one promotion. Forwardfuture.ai. Forwardfuture.ai. I'm going to drop it in chat right now. We are putting out incredible original content. In fact, I'm going to show it to you. Can y'all see this now? Okay, we have original content. We have authors coming on board writing amazing essays for us. Here's one from Tomas Hernando Kaufman about model routing. We have one from our very own Jonah. We have Ahmad Osman about local inference. Scott Santins, UBI expert, talking about UBI. Um, uh, Dave Shapiro been on. I mean, please go to forwardfuture.ai, subscribe, check it out. We are going to be publishing incredible original pieces. Go check out the Discord. Thank you, Brian. Check out the Discord. We spun up the Discord. Brian, uh, producer, editor Brian on our team is managing it. It is a great place to come hang out, talk to us, talk to other AI enthusiasts. Uh, and thank you again. Not Ben Shapiro, Dave Shapiro. Um, any chance for Oh, okay. Any chance for Sim City? Let's see. Did that even finish? Okay. Okay. I guess we're extending stream just for another moment. I don't know. I don't know. I turn two. Did it finish? No, I can't tell if it's finished or not. Okay. So, no, it's not done. It's been going for 22 minutes. 100,000 tokens. Um, when it's done, I'll publish it. I'll maybe tweet about it. So, if you're not following me on Twitter, Matthew Berman, go follow me and if it's finished, I'll publish it and I'll post it there. All right. Thanks everybody. I will see you next time. Fun day.