WEBVTT 00:00:02.239 --> 00:00:04.780 Welcome to the AI Chat Podcast. I'm your host, 00:00:04.860 --> 00:00:07.179 Jaden Schaefer. Every day I cover cutting -edge 00:00:07.179 --> 00:00:09.320 AI news and talk with the leaders behind it, 00:00:09.339 --> 00:00:11.300 breaking down what it means for your life and 00:00:11.300 --> 00:00:18.039 business. Anthropic is launching Claude Science. 00:00:18.219 --> 00:00:20.739 This is going to be a workbench for researchers. 00:00:21.179 --> 00:00:24.100 Base44 is launching their very own AI model. 00:00:24.239 --> 00:00:26.239 They're trying to protect their $100 million 00:00:26.239 --> 00:00:29.440 in annual recurring revenue for their vibe coding 00:00:29.440 --> 00:00:33.560 business. And X has just launched a hosted MCP 00:00:33.560 --> 00:00:36.079 server. So it's going to open its API to Claude, 00:00:36.079 --> 00:00:38.460 Cursor, Grok, Build, which if you know anything 00:00:38.460 --> 00:00:41.439 about the backstory of X and their API, this 00:00:41.439 --> 00:00:46.240 is quite a big deal. Claude Sonnet 5. It's going 00:00:46.240 --> 00:00:49.579 to be $2 per million input tokens. And Google 00:00:49.579 --> 00:00:52.920 is shipping Nano Banana 2 Lite and Gemini Omni 00:00:52.920 --> 00:00:55.679 Flash 2 developers. So tons coming out from basically 00:00:55.679 --> 00:00:58.329 every top AI model today. We're going to get 00:00:58.329 --> 00:01:00.250 into all of that. Hopefully, it's not going to 00:01:00.250 --> 00:01:03.909 be too noisy from my end. It is a hot, sunny 00:01:03.909 --> 00:01:05.750 day in North Carolina where I'm at right now, 00:01:05.810 --> 00:01:08.790 and I am currently bouncing a baby on my lap. 00:01:08.950 --> 00:01:11.950 My wife did not sleep a lot last night, so trying 00:01:11.950 --> 00:01:14.129 to let her get some rest, holding the baby. You 00:01:14.129 --> 00:01:15.989 probably heard me say this before on the podcast. 00:01:16.209 --> 00:01:18.890 But anyways, if you can hear a little gurgling 00:01:18.890 --> 00:01:21.170 and gooing in the background, that would be what 00:01:21.170 --> 00:01:23.939 it's from. If there's any repetitive AI tasks 00:01:23.939 --> 00:01:27.000 that you do, like writing newsletters or emails 00:01:27.000 --> 00:01:29.700 or generating graphics or anything you do on 00:01:29.700 --> 00:01:31.640 a repeated basis, I would love for you to go 00:01:31.640 --> 00:01:35.439 check out my startup, AIbox .ai. We have a no 00:01:35.439 --> 00:01:37.980 -code AI app builder where you basically just 00:01:37.980 --> 00:01:40.540 explain what you want it to do, whatever your 00:01:40.540 --> 00:01:42.799 workflow is, and it automatically links together 00:01:42.799 --> 00:01:45.000 different AI models and creates a workflow for 00:01:45.000 --> 00:01:47.959 you that you can use on a repeated basis. This 00:01:47.959 --> 00:01:49.700 saves me a lot of time because AI definitely 00:01:49.700 --> 00:01:53.280 is a huge... for being more productive. But at 00:01:53.280 --> 00:01:54.959 the same time, if you're just using it to do 00:01:54.959 --> 00:01:56.640 the same thing over and over every single day, 00:01:56.739 --> 00:01:58.819 you should probably just automate that. So if 00:01:58.819 --> 00:02:01.140 you want to give it a try, go to AI box dot AI 00:02:01.140 --> 00:02:03.599 slash builder. I'll leave a link in the description 00:02:03.599 --> 00:02:05.799 to go check out the builder platform that we 00:02:05.799 --> 00:02:08.259 have. We have over 80 different AI models that 00:02:08.259 --> 00:02:10.139 you can link together and build some really incredible 00:02:10.139 --> 00:02:12.300 automations to hopefully save you a ton of time. 00:02:12.340 --> 00:02:14.819 And it's only $8 .99 a month to get started. 00:02:14.919 --> 00:02:16.659 I'll leave a link in the description. Kicking 00:02:16.659 --> 00:02:18.939 it off with Anthropic. They've just launched 00:02:18.939 --> 00:02:20.860 Claude Science. So this is going to be a workbench. 00:02:20.979 --> 00:02:23.580 It's connecting 60 different scientific databases 00:02:23.580 --> 00:02:26.060 and a bunch of pre -built research tools for 00:02:26.060 --> 00:02:29.039 genomics, protein structures, chemistry. And 00:02:29.039 --> 00:02:31.139 all of this is going to run on Claude Opus 4 00:02:31.139 --> 00:02:34.400 .8. So sorry, Fable 5, not there yet. I'm sure 00:02:34.400 --> 00:02:36.639 when that comes out, we may get that rolled in 00:02:36.639 --> 00:02:39.439 as well. And it's going to come with $30 ,000 00:02:39.439 --> 00:02:41.539 in free credits. They're going to be giving this 00:02:41.539 --> 00:02:44.099 way to 50 different academic projects. And they're 00:02:44.099 --> 00:02:46.680 making this bet that basically workflow is going 00:02:46.680 --> 00:02:49.830 to beat raw capacity for getting these... researchers 00:02:49.830 --> 00:02:53.590 to be using Anthropic. So there's no new model 00:02:53.590 --> 00:02:56.310 or gating, but Claude Science uses the same Claude 00:02:56.310 --> 00:02:58.889 Opus 4 .8. It's basically available to everybody. 00:02:59.009 --> 00:03:01.770 And there is no biology fine -tuning or specialized 00:03:01.770 --> 00:03:03.750 weights. They've been testing this with a few 00:03:03.750 --> 00:03:05.490 different people and have had some impressive 00:03:05.490 --> 00:03:09.189 wins. They had UCSF Brain Tumor Center that was 00:03:09.189 --> 00:03:12.250 using it to compress geloma germline analysis 00:03:12.250 --> 00:03:15.449 workflows. We had the Allen Institute's Jerome 00:03:15.449 --> 00:03:18.210 Lecoq, who is building a computational review 00:03:18.210 --> 00:03:20.530 pipeline. line. And we have Novo Nordisk, who's 00:03:20.530 --> 00:03:22.270 one of their launch partners, what it's actually 00:03:22.270 --> 00:03:24.370 doing is that it's able to spin up these multi 00:03:24.370 --> 00:03:26.430 agent architectures. And they're going to let 00:03:26.430 --> 00:03:29.009 all of these scientific researchers create parallel 00:03:29.009 --> 00:03:31.289 sub assistants. And they're going to be doing 00:03:31.289 --> 00:03:33.289 sequence analysis, they're going to be doing 00:03:33.289 --> 00:03:35.030 structural prediction, they're going to be doing 00:03:35.030 --> 00:03:36.930 fact checking. And the reason why this is so 00:03:36.930 --> 00:03:38.930 powerful is because all of that can be happening 00:03:38.930 --> 00:03:41.669 while they're also keeping their full reproducibility. 00:03:41.770 --> 00:03:44.090 So the code, the environments, the message history, 00:03:44.289 --> 00:03:45.949 all of that, they're going to be able to reproduce. 00:03:46.050 --> 00:03:48.879 So there's a bunch of basically custom that scientists 00:03:48.879 --> 00:03:51.319 need. And Anthropic has built that in. They're 00:03:51.319 --> 00:03:53.780 giving away a bunch of grants, hoping to show 00:03:53.780 --> 00:03:55.419 people that this is the, you know, Anthropic 00:03:55.419 --> 00:03:58.719 is the best place if you're doing science. Okay, 00:03:58.819 --> 00:04:01.580 Base44 is launching their very own AI model. 00:04:01.780 --> 00:04:04.120 And the reason why is, by the way, I didn't realize, 00:04:04.280 --> 00:04:07.280 but Base44 is owned by Wix. It's an incredible 00:04:07.280 --> 00:04:09.180 vibe coding platform. I've been really impressed 00:04:09.180 --> 00:04:11.379 with it. It's not the one that I primarily use 00:04:11.379 --> 00:04:13.300 because I'm just using cloud code for mostly 00:04:13.300 --> 00:04:15.020 everything I do. But if you're getting started, 00:04:15.139 --> 00:04:17.279 I think anywhere between Loveable and Base44 00:04:17.259 --> 00:04:19.519 Base44, those are my two favorite. Anywhere is 00:04:19.519 --> 00:04:21.899 also a great platform. But in any case, Base44 00:04:21.899 --> 00:04:24.660 is $100 million in annual recurring revenue, 00:04:24.819 --> 00:04:26.759 and they just launched their brand new model 00:04:26.759 --> 00:04:29.100 called Base1. So this is their very own LLM. 00:04:29.120 --> 00:04:30.459 This is actually very similar to what Cursor 00:04:30.459 --> 00:04:32.720 is doing, by the way. But this LLM is trained 00:04:32.720 --> 00:04:35.199 on tens of millions of real user interactions. 00:04:35.220 --> 00:04:37.560 So what's interesting is because they're able 00:04:37.560 --> 00:04:39.899 to... you know, have tens of millions of users 00:04:39.899 --> 00:04:42.500 using them to create their tools, they can take 00:04:42.500 --> 00:04:44.540 all of those conversations. And I'm not sure 00:04:44.540 --> 00:04:46.120 if it's every single user, if they had people 00:04:46.120 --> 00:04:48.339 opt in, it's probably every single user, and 00:04:48.339 --> 00:04:49.899 it's just part of their terms of service. But 00:04:49.899 --> 00:04:51.339 they're taking everything that those people have 00:04:51.339 --> 00:04:53.579 been saying, and they're training their model 00:04:53.579 --> 00:04:55.639 to basically fine tune what people need when 00:04:55.639 --> 00:04:58.100 they're building websites, apps, and software. 00:04:58.220 --> 00:05:00.480 Now, what's interesting to me about this is they 00:05:00.480 --> 00:05:02.639 know exactly like when someone's building software, 00:05:02.759 --> 00:05:05.139 and you know, theoretically, anthropic and codex, 00:05:05.139 --> 00:05:06.720 like for opening, I should have this as well. 00:05:06.740 --> 00:05:09.100 But they know every time that someone asks it 00:05:09.100 --> 00:05:11.180 to build something and it does it, and then they 00:05:11.180 --> 00:05:12.839 give a follow -up, they're like, oh shoot, every 00:05:12.839 --> 00:05:14.740 time someone asks for this, you know, for X, 00:05:14.839 --> 00:05:17.420 Y, Z, they typically ask for these next five 00:05:17.420 --> 00:05:19.920 things. And that could either mean that when 00:05:19.920 --> 00:05:21.720 they first ask, they don't know exactly what 00:05:21.720 --> 00:05:23.620 they want, or there's traditionally a bug that 00:05:23.620 --> 00:05:25.759 follows, or there's a gotcha or something like 00:05:25.759 --> 00:05:28.579 that. But essentially they can fine tune this 00:05:28.579 --> 00:05:31.660 model that they have to know exactly what they 00:05:31.660 --> 00:05:34.399 want to, what people are basically, the direction 00:05:34.399 --> 00:05:35.939 they're going before they even go there, right? 00:05:35.959 --> 00:05:37.240 They're like, hey, I want like a fit. in this 00:05:37.240 --> 00:05:38.839 app and like oh these are probably the five features 00:05:38.839 --> 00:05:40.740 they want so either they'll ask them right away 00:05:40.740 --> 00:05:42.459 or they'll just start building stuff in or they'll 00:05:42.459 --> 00:05:44.379 set up the architecture in a way that is ready 00:05:44.379 --> 00:05:46.379 to go now there's a couple cool things that they 00:05:46.379 --> 00:05:48.639 can do with these fine -tuned models one of them 00:05:48.639 --> 00:05:50.720 being that they can make them much more efficient 00:05:50.720 --> 00:05:52.779 and so they're able to cut the costs for their 00:05:52.779 --> 00:05:55.639 users which of course is fantastic but also if 00:05:55.639 --> 00:05:57.360 someone's talking to their model and they're 00:05:57.360 --> 00:05:59.959 like oh you know what our you know our base one 00:05:59.959 --> 00:06:01.600 model would be a lot better than sending it to 00:06:01.600 --> 00:06:05.879 opus 4 .8 all of a sudden they're able to switch 00:06:05.879 --> 00:06:07.860 people over to the more efficient model, save 00:06:07.860 --> 00:06:11.019 the user money, and they're also keeping exclusive 00:06:11.019 --> 00:06:13.180 data, right? Because if they're not sending data 00:06:13.180 --> 00:06:17.750 over to OpenAI or Gemini or... all of a sudden 00:06:17.750 --> 00:06:19.810 they're not able to train with all of that data. 00:06:19.850 --> 00:06:23.009 So we're also seeing the same thing coming out 00:06:23.009 --> 00:06:25.790 of Cursor. So a bunch of different players that 00:06:25.790 --> 00:06:28.470 are interacting with code are basically intercepting 00:06:28.470 --> 00:06:30.250 the code and the conversations people are having 00:06:30.250 --> 00:06:32.410 and knowing what people want, and they're being 00:06:32.410 --> 00:06:34.670 able to train their own fine -tuned models with 00:06:34.670 --> 00:06:37.269 it. So base one, basically they're saying this 00:06:37.269 --> 00:06:40.050 is designed to replace Anthropix Opus for app 00:06:40.050 --> 00:06:42.350 generation workloads, and it's going to try to 00:06:42.350 --> 00:06:44.430 give them a really direct control over latency, 00:06:44.610 --> 00:06:46.899 cost, and inference speed. And they're not going 00:06:46.899 --> 00:06:48.620 to have to rely on any third -party APIs. And 00:06:48.620 --> 00:06:50.579 also, by the way, you know, this is a big cost 00:06:50.579 --> 00:06:52.060 for them because when you get a subscription 00:06:52.060 --> 00:06:54.720 to Base44, you pay them for tokens, but a huge 00:06:54.720 --> 00:06:56.720 chunk of your token bill just goes straight to 00:06:56.720 --> 00:06:58.860 Anthropic, right? It's like they're given a huge 00:06:58.860 --> 00:07:00.620 chunk, but if they can do it directly on their 00:07:00.620 --> 00:07:03.199 servers and if they have, you know, a cost -efficient 00:07:03.199 --> 00:07:04.839 way to do that, they're actually able to make 00:07:04.839 --> 00:07:07.480 more money. So Lovable that I've mentioned many 00:07:07.480 --> 00:07:10.500 times is, of course, their rival in Sweden, and 00:07:10.500 --> 00:07:12.980 they still use external LLMs. They have about 00:07:12.980 --> 00:07:15.639 $500 billion in AR. so they're about five times 00:07:15.639 --> 00:07:17.839 bigger. But they do have some really solid backing 00:07:17.839 --> 00:07:20.139 because Wix acquired them for about $80 million 00:07:20.139 --> 00:07:23.259 when they were only six months old with eight 00:07:23.259 --> 00:07:25.439 employees. Their parent company is now cutting 00:07:25.439 --> 00:07:28.720 about... 20 % of their workforce. So, you know, 00:07:28.740 --> 00:07:31.139 Wix is cutting down, but evidently Base44 is 00:07:31.139 --> 00:07:33.740 accelerating. And my prediction here is that 00:07:33.740 --> 00:07:36.839 if Base44 really pulls this off, Base44 will 00:07:36.839 --> 00:07:38.639 actually end up being much bigger than Wix, which 00:07:38.639 --> 00:07:40.139 is pretty wild considering they bought it for 00:07:40.139 --> 00:07:42.800 $80 million. It's already doing more than $100 00:07:42.800 --> 00:07:45.379 million in annual recurring revenue. So they 00:07:45.379 --> 00:07:47.519 were just, you know, blasted past even the acquisition 00:07:47.519 --> 00:07:50.639 price. This was an amazing deal for Wix. All 00:07:50.639 --> 00:07:52.199 right. The next thing we got to talk about is 00:07:52.199 --> 00:07:55.339 that X has just launched a hosted MCP service. 00:07:55.339 --> 00:07:57.939 server and if you know x like they when they 00:07:57.939 --> 00:08:00.660 when elon originally bought twitter he immediately 00:08:00.660 --> 00:08:03.379 shut down the api because i think open ai and 00:08:03.379 --> 00:08:05.000 a bunch of other people just had like a full 00:08:05.000 --> 00:08:08.420 firehose blast of all of the api data and they 00:08:08.420 --> 00:08:10.199 were using it for training elon shut that down 00:08:10.199 --> 00:08:12.100 realizing how valuable it was we saw similar 00:08:12.100 --> 00:08:14.019 moves from reddit and other players afterwards 00:08:14.019 --> 00:08:17.240 but now that x is kind of launching this new 00:08:17.240 --> 00:08:20.800 mcp they're going to let tools like claude cursor 00:08:20.800 --> 00:08:26.240 grok all tap into x ai's API, and specifically 00:08:26.240 --> 00:08:28.959 your own account. So if you have your own account 00:08:28.959 --> 00:08:31.160 or your own permissions, you can let the you 00:08:31.160 --> 00:08:33.600 can let the MCP tap into that. I'm sure there's 00:08:33.600 --> 00:08:36.200 a ton of like market sentiment analysis, you 00:08:36.200 --> 00:08:37.840 know, great things that you can do with this, 00:08:37.899 --> 00:08:39.879 there is an API. So big firms, you know, like 00:08:39.879 --> 00:08:42.200 hedge funds that are trying to, you know, measure 00:08:42.200 --> 00:08:44.399 people's sentiment analysis on the FIFA World 00:08:44.399 --> 00:08:46.320 Cup to determine how many hot dogs are going 00:08:46.320 --> 00:08:48.019 to be sold and blah, blah, blah, blah, and do 00:08:48.019 --> 00:08:50.379 all their like crazy, you know, hedge fund calculations, 00:08:50.679 --> 00:08:53.179 those types of things, they already have an API 00:08:53.179 --> 00:08:54.929 there. going to be using this. This is more for 00:08:54.929 --> 00:08:57.389 your average user. An MCP means that you can 00:08:57.389 --> 00:09:00.450 get all of XAI's data into your cloud that you're 00:09:00.450 --> 00:09:02.950 actively using. So for regular people, I personally 00:09:02.950 --> 00:09:05.389 would love to connect the MCP for when I talk 00:09:05.389 --> 00:09:07.789 about AI news here, I would love to get maybe 00:09:07.789 --> 00:09:09.789 like some sort of brief on every story I'm going 00:09:09.789 --> 00:09:12.370 to cover with the top tweets and the top comments 00:09:12.370 --> 00:09:14.509 on those tweets. So if my MCP could go and pull 00:09:14.509 --> 00:09:16.549 that, that'd be really useful. Traditionally 00:09:16.549 --> 00:09:19.309 trying to get like cloud to go and scrape X for 00:09:19.309 --> 00:09:22.289 good tweets is notoriously hard. I've had a hard 00:09:22.289 --> 00:09:24.090 time doing that. And I basically just have to 00:09:24.090 --> 00:09:26.909 go manually scroll X myself. It would be fantastic 00:09:26.909 --> 00:09:28.570 if I could save some time and just, you know, 00:09:28.570 --> 00:09:31.190 surface the best things. Basically, I mean, what's 00:09:31.190 --> 00:09:32.789 kind of cool is if you have an MCP into something 00:09:32.789 --> 00:09:34.549 like X, and I know I'm getting off the rails 00:09:34.549 --> 00:09:36.230 here, but you can imagine the same thing with 00:09:36.230 --> 00:09:38.289 Reddit or any other social platform. You can 00:09:38.289 --> 00:09:41.509 fine tune your own algorithm, which I think threads 00:09:41.509 --> 00:09:43.190 and Mark Zuckerberg's trying to let people kind 00:09:43.190 --> 00:09:45.909 of do that with threads. But it's like in threads 00:09:45.909 --> 00:09:47.929 anyways, it's like show more posts about this 00:09:47.929 --> 00:09:49.750 specific topic for the next like four weeks. 00:09:49.830 --> 00:09:51.649 And then anyways, it's not like a permanent thing. 00:09:52.200 --> 00:09:54.440 This would be cool. I would honestly almost remake 00:09:54.440 --> 00:09:56.879 my own X feed because I probably get, I'm sure 00:09:56.879 --> 00:09:58.360 I interact with it, but I feel like I get like 00:09:58.360 --> 00:10:00.860 way too much politics in my thread that I really 00:10:00.860 --> 00:10:03.220 want to interact with in any given day. So, and 00:10:03.220 --> 00:10:05.720 it's probably my own fault, but it would be really 00:10:05.720 --> 00:10:07.580 cool to just intentionally be able to use this 00:10:07.580 --> 00:10:09.980 MCP on X and say, look, I just want AI news. 00:10:10.100 --> 00:10:12.100 I just want, you know, news from these, you know, 00:10:12.100 --> 00:10:14.620 500 companies that I'm trying to follow in AI 00:10:14.620 --> 00:10:18.740 and get more of a streamlined stream of content 00:10:18.740 --> 00:10:22.389 around that. That'd be cool for me. you know, 00:10:22.409 --> 00:10:24.450 the first big company to do an MCP. It feels 00:10:24.450 --> 00:10:26.649 like basically everyone has an MCP, even my own 00:10:26.649 --> 00:10:29.970 company, AIbox .ai. We have an MCP. So, you know, 00:10:29.970 --> 00:10:31.889 traditionally people would go to our site. There's 00:10:31.889 --> 00:10:34.190 80 different AI models, audio, text, video, and 00:10:34.190 --> 00:10:35.649 people would go chat with them in our playground. 00:10:35.769 --> 00:10:37.830 And, you know, you pay like $9 a month and you 00:10:37.830 --> 00:10:41.149 get access to every model. But a lot of people 00:10:41.149 --> 00:10:43.250 are like, look, all of my workload is, all of 00:10:43.250 --> 00:10:45.610 my work is happening inside of Clod or inside 00:10:45.610 --> 00:10:47.809 of ChatGPT. So we created an MCP and now you 00:10:47.809 --> 00:10:49.870 can get access to all of our different tools. 00:10:49.990 --> 00:10:51.889 You can get, you know, like Clod can, have image 00:10:51.889 --> 00:10:54.309 generation and audio generation, which it didn't 00:10:54.309 --> 00:10:58.110 have before. So MCPs are very valuable and XAI 00:10:58.110 --> 00:11:00.830 is not the first one. GitHub, Slack, Stripe, 00:11:00.830 --> 00:11:03.230 Salesforce, all of them have official MCP endpoints. 00:11:03.330 --> 00:11:05.990 The API pricing, if you're going to do that with 00:11:05.990 --> 00:11:09.789 X is still going to be 0 .015 per published post 00:11:09.789 --> 00:11:12.929 and about... 20 cents per post with links. That 00:11:12.929 --> 00:11:14.809 was set earlier this year. They basically said 00:11:14.809 --> 00:11:17.389 they're trying to curb spam at scale. But the 00:11:17.389 --> 00:11:20.190 hosted server is showing some new capabilities. 00:11:20.350 --> 00:11:22.250 There's search, there's post retrieval, there's 00:11:22.250 --> 00:11:24.769 user lookup, there's trend analysis. All of that 00:11:24.769 --> 00:11:27.509 is cutting down a lot of integration time for 00:11:27.509 --> 00:11:29.710 people trying to work with X. So I'm going to 00:11:29.710 --> 00:11:32.950 be excited to test that out. Anthropic is shipping 00:11:32.950 --> 00:11:36.250 Claude Sonnet 5. It's going to be $2 per million 00:11:36.250 --> 00:11:39.169 input tokens, which is undercutting Opus 4 .8. 00:11:39.269 --> 00:11:42.940 And it is about... 63 .2 % on agentic coding. 00:11:43.159 --> 00:11:46.799 So it's within about six points of Opus 4 .8. 00:11:46.860 --> 00:11:48.840 You can only imagine, right, the next model, 00:11:48.940 --> 00:11:51.139 which is Fable 5. I mean, they already dropped 00:11:51.139 --> 00:11:52.600 it, but it's going to be significantly better. 00:11:52.679 --> 00:11:54.440 It's interesting because the, you know, their 00:11:54.440 --> 00:11:56.399 headline model got pulled back. So now they just 00:11:56.399 --> 00:11:59.419 have all of their like, all of their their worst 00:11:59.419 --> 00:12:00.940 models are, you know, they're more efficient 00:12:00.940 --> 00:12:03.700 models. They're less powerful models, they're 00:12:03.700 --> 00:12:05.220 starting to catch up with their main model, because 00:12:05.220 --> 00:12:06.820 they can't even, you know, update their main 00:12:06.820 --> 00:12:08.620 model to be its best, which is really interesting 00:12:08.620 --> 00:12:11.029 to me. So this model is going to basically become 00:12:11.029 --> 00:12:13.830 the default for all free and pro users starting 00:12:13.830 --> 00:12:16.230 on Tuesday. I think that is just showing that 00:12:16.230 --> 00:12:19.169 agent capable AI is now really important across 00:12:19.169 --> 00:12:21.110 basically every tier. This isn't going to be 00:12:21.110 --> 00:12:23.690 just a premium feature. And I do think that it's 00:12:23.690 --> 00:12:27.629 promotional pricing, the $2. for the inputs and 00:12:27.629 --> 00:12:29.470 the $10 for the outputs. That is going to be 00:12:29.470 --> 00:12:31.529 running through August 31st. And we see this 00:12:31.529 --> 00:12:33.169 from Anthropic a lot. Like I think if I go on 00:12:33.169 --> 00:12:35.029 my account right now, they're like, you know, 00:12:35.049 --> 00:12:38.090 the beginning of July, it's 2x credit usage. 00:12:38.269 --> 00:12:39.809 And I mean, basically at this point, I'm just 00:12:39.809 --> 00:12:42.730 trying to get my max out my usage every single 00:12:42.730 --> 00:12:44.929 week because I feel like you're getting a really 00:12:44.929 --> 00:12:46.529 great deal. And I don't know how long that will 00:12:46.529 --> 00:12:49.429 last, but that's going to increase from $2 to 00:12:49.429 --> 00:12:52.490 $3, which is still cheaper than Opus 4 .8 and 00:12:52.490 --> 00:12:55.639 GPT 5 .5 and Gemini 3 .1 Pro. So it's still cheaper 00:12:55.639 --> 00:12:58.580 but it's uh yeah until august 31st we got the 00:12:58.580 --> 00:13:00.279 two dollar price it's going to then go up to 00:13:00.279 --> 00:13:02.600 three dollars um and you know maybe we'll have 00:13:02.600 --> 00:13:04.840 new models coming out then so who knows what 00:13:04.840 --> 00:13:07.519 happens in the future but on knowledge work benchmarks 00:13:07.519 --> 00:13:11.899 sonnet 5 is slightly better than opus 4 .8 which 00:13:11.899 --> 00:13:14.919 is pretty crazy right like sonnet is the you 00:13:14.919 --> 00:13:17.340 know the the lower quality model than opus but 00:13:17.340 --> 00:13:19.840 because they can't upgrade opus now the worst 00:13:19.840 --> 00:13:23.820 version is actually better than opus 4 .8 i think 00:13:23.820 --> 00:13:25.500 that's just kind of showing though the cost per 00:13:25.500 --> 00:13:28.600 task efficiency is now really important it's 00:13:28.600 --> 00:13:30.820 more than more important than maybe the raw reasoning 00:13:30.820 --> 00:13:33.360 rank and if you want to spend a ton of money 00:13:33.360 --> 00:13:35.179 you can go use something like clod code and turn 00:13:35.179 --> 00:13:37.759 on like you know their super code mode or whatever 00:13:38.409 --> 00:13:41.070 It gives you, you know, waste more reasoning, 00:13:41.169 --> 00:13:43.169 but it uses a lot more tokens. A lot of early 00:13:43.169 --> 00:13:46.429 testers say that there are fewer mid -task stalls 00:13:46.429 --> 00:13:49.330 in some of the multi -step workflows. Zapier's 00:13:49.330 --> 00:13:51.529 two -part Salesforce automation completed where 00:13:51.529 --> 00:13:53.649 I think a bunch of previous models, they were 00:13:53.649 --> 00:13:55.870 trying to do that two -part Salesforce automation, 00:13:56.110 --> 00:13:58.129 and it was getting about halfway through and 00:13:58.129 --> 00:14:00.350 it abandoned it. So anyways, it seems like this 00:14:00.350 --> 00:14:02.529 is going to be much more efficient. Speaking 00:14:02.529 --> 00:14:04.330 of new models, they're not the only one. Google 00:14:04.330 --> 00:14:07.090 is shipping Nano Banana 2 Lite and also Gemini 00:14:07.090 --> 00:14:10.639 Omni Flash. to developers. It is a text -to -image 00:14:10.639 --> 00:14:12.879 model. It's going to generate in about four seconds, 00:14:12.899 --> 00:14:17.779 and it costs 0 .034 cents for a 1K image. Gemini 00:14:17.779 --> 00:14:20.419 OmniFlash, which is a video generation model, 00:14:20.539 --> 00:14:24.159 is about 10 cents per second of output, which 00:14:24.159 --> 00:14:27.620 honestly is not bad. These things can be insanely 00:14:27.620 --> 00:14:29.940 expensive. We literally had OpenAI that canceled 00:14:29.940 --> 00:14:32.700 Sora and is discontinuing it because it was just 00:14:32.700 --> 00:14:35.860 so expensive. So we had the capability to do 00:14:35.860 --> 00:14:37.669 AI video. It's just really expensive. So the 00:14:37.669 --> 00:14:39.690 fact that Gemini is getting that down to 10 cents 00:14:39.690 --> 00:14:42.309 per second of output, not bad. You know, if you 00:14:42.309 --> 00:14:44.970 were doing a one hour long video, it's going 00:14:44.970 --> 00:14:47.019 to be pretty expensive. But most of the videos 00:14:47.019 --> 00:14:48.720 that these are generating are like five seconds 00:14:48.720 --> 00:14:51.080 long. So that's 50 cents a video, which I know 00:14:51.080 --> 00:14:53.080 actually sounds kind of expensive. If it's 10 00:14:53.080 --> 00:14:55.240 second video, it's a dollar. If it's a, you know, 00:14:55.240 --> 00:14:58.539 100 second video, it's $10. But you know, that's 00:14:58.539 --> 00:15:01.539 that's the cost. Okay, nano banana to light is 00:15:01.539 --> 00:15:03.480 going to be replacing the original nano banana, 00:15:03.659 --> 00:15:05.480 it's going to be the recommend is the recommended 00:15:05.480 --> 00:15:07.639 tier. And I think they're saying that it's like 00:15:07.639 --> 00:15:09.419 if you have a prompt that worked really well 00:15:09.419 --> 00:15:11.340 for the original nano banana, it's going to still 00:15:11.340 --> 00:15:14.299 work for nano banana to light, which is great. 00:15:14.340 --> 00:15:15.899 I hate it when the model changes and the prompts 00:15:15.899 --> 00:15:18.720 are kind of get off. They said there's going 00:15:18.720 --> 00:15:21.340 to be character consistency, they have really 00:15:21.340 --> 00:15:23.659 aggressive speed cost optimization that they've 00:15:23.659 --> 00:15:25.279 applied. So it's kind of interesting, right? 00:15:25.320 --> 00:15:27.980 It's like two, it's nano banana to light, like 00:15:27.980 --> 00:15:29.720 this is supposed to be like a super fast, super 00:15:29.720 --> 00:15:32.100 light version. And it's they're trying to get 00:15:32.100 --> 00:15:33.820 this thing to still be comparable to nano banana, 00:15:34.019 --> 00:15:36.100 but way faster and way cheaper. And I'm really 00:15:36.100 --> 00:15:37.759 excited when we do these kind of optimizations 00:15:37.759 --> 00:15:40.500 on the image models. They also have Omni flash, 00:15:40.659 --> 00:15:44.220 which is going to match VO three, one fast pricing 00:15:44.220 --> 00:15:46.759 while it's also adding conversational editing 00:15:46.759 --> 00:15:49.500 multimodal input mixing so they could do text 00:15:49.500 --> 00:15:51.779 image and video and it's going to do text to 00:15:51.779 --> 00:15:54.700 action synchronization for any sort of like graphics 00:15:54.700 --> 00:15:56.580 that you have on the screen which i think is 00:15:57.309 --> 00:15:59.309 So for videographers, this is actually really 00:15:59.309 --> 00:16:02.409 cool. Nano Banana 2 Lite is going to ship across 00:16:02.409 --> 00:16:04.690 nine of the different products that Google already 00:16:04.690 --> 00:16:06.230 has. So it's going to be inside of AI Studio. 00:16:06.549 --> 00:16:08.509 It's going to be inside of the Gemini API. It's 00:16:08.509 --> 00:16:10.690 going to be inside of AI Mode in Search. So when 00:16:10.690 --> 00:16:12.330 you're on Google just searching, it's going to 00:16:12.330 --> 00:16:14.570 be on the Gemini app, Notebook LM, Google Photos, 00:16:14.570 --> 00:16:16.789 Stitch, Google Flow, and then it's also inside 00:16:16.789 --> 00:16:19.250 of Google Ads, right? They got to make money 00:16:19.250 --> 00:16:21.509 from the ads off of all of this. I'm excited 00:16:21.509 --> 00:16:23.250 to see what Google comes out with. I'm excited 00:16:23.250 --> 00:16:24.809 for all the new models today. Guys, thank you 00:16:24.809 --> 00:16:26.899 so much for tuning into the podcast. make sure 00:16:26.899 --> 00:16:29.759 if you want to check out the mcp for ai box that 00:16:29.759 --> 00:16:33.000 you go check out ai box .ai mcp basically you're 00:16:33.000 --> 00:16:35.600 going to get all 80 of the top AI models right 00:16:35.600 --> 00:16:38.240 inside of Claude or Gemini or ChatGPT, whatever 00:16:38.240 --> 00:16:40.379 you work with the most. You can get all of the 00:16:40.379 --> 00:16:42.220 other models inside of there so you can actually 00:16:42.220 --> 00:16:44.720 call them. Really, really useful. And as always, 00:16:44.759 --> 00:16:46.360 if you want to get all of these news stories 00:16:46.360 --> 00:16:48.019 that I talk about on the podcast straight into 00:16:48.019 --> 00:16:51.720 your inbox every day, go to AIChatDaily .com. 00:16:51.840 --> 00:16:53.960 That's my website that goes along with this podcast. 00:16:54.200 --> 00:16:55.639 There's a big subscribe button you can hit in 00:16:55.639 --> 00:16:58.259 the top corner. And if you hit that, you can 00:16:58.259 --> 00:16:59.940 put in your email and I'll send you all of these 00:16:59.940 --> 00:17:03.059 stories and more in -depth articles. where it 00:17:03.059 --> 00:17:04.660 breaks down all of the information. You can see 00:17:04.660 --> 00:17:06.400 all of that as well. Thank you so much for tuning 00:17:06.400 --> 00:17:08.339 into the podcast, guys. I hope you have a fantastic 00:17:08.339 --> 00:17:09.880 day and I'll catch you in the next episode.