Anthropic launches Claude Science and a new model

**SPEAKER_1** (0:00)
Anthropic is launching Claude Science. This is gonna be a workbench for researchers. Base 44 is launching their very own AI model. They're trying to protect their $100 million in annual recurring revenue for their vibe coding business. And X has just launched a hosted MCP server. So it's gonna open its API to Claude Cursor Grok Build, which if you know anything about the backstory of X and their API, this is quite a big deal. Anthropic is shipping Claude Sonnet 5 It's gonna be $2 per million input tokens. And Google is shipping Nano Banana 2 Lite and Gemini Omni Flash 2 developers. So tons coming out from basically every top AI model today. We're gonna get into all of that. Hopefully it's not gonna be too noisy from my end. It is a hot sunny day in North Carolina where I'm at right now. And I am currently bouncing a baby on my lap. My wife did not sleep a lot last night. So trying to let her get some rest holding the baby. You probably heard me say this before on the podcast. But anyways, if you can hear a little gurgling and gooing in the background, that would be what it's from. If there's any repetitive AI tasks that you do, like writing newsletters or emails or generating graphics or anything you do on a repeated basis, I would love for you to go check out my startup AI box.ai. We have a no code AI app builder where you basically just explain what you want it to do, whatever your workflow is, and automatically links together different AI models and creates a workflow for you that you can use on a repeated basis. This saves me a lot of time because AI definitely is a huge unlock for being more productive. But at the same time, if you're just using it to do the same thing over and over every single day, you should probably just automate that. So if you want to give it a try, go to aibox.ai/builder.
I'll leave a link in the description to go check out the builder platform that we have. We have over 80 different AI models that you can link together and build some really incredible automations that will hopefully save you a ton of time. And it's only $8.99 a month to get started. I'll leave a link in the description. Kicking it off with Anthropic, they have just launched Claude Science. So this is going to be a workbench. It's connecting 60 different scientific databases and a bunch of pre-built research tools for genomics, protein structures, chemistry. And all of this is going to run on Claude Opus 4.8. So sorry, Fable 5 not there yet. I'm sure when that comes out, we may get that rolled in as well. And it's going to come with $30,000 in free credits. They're going to be giving this way to 50 different academic projects. And they're making this bet that basically workflow is going to be raw capacity for getting these researchers to be using Anthropic. So there's no new model or gating, but Claude Science uses the same Claude Opus 4.8. It's basically available to everybody. And there is no biology fine tuning or specialized weights. They've been testing this with a few different people and have had some impressive wins. They had UCSF Brain Tumor Center that was using it to compress geloma germline analysis workflows. We had the Allen Institutes, Jerome Lecoq, who is building a computational review pipeline. And we have Novo Nordisk, who is one of their launch partners. What it's actually doing is that it's able to spin up these multi-agent architectures, and they're going to let all of these scientific researchers create parallel sub-assistance, and they're going to be doing sequence analysis. They're going to be doing structural prediction. They're going to be doing fact checking. And the reason why this is so powerful is because all of that can be happening while they're also keeping their full reproducibility. So the code, the environments, the message history, all of that they're going to be able to reproduce. So there's a bunch of basically custom tools that scientists need, and Anthropic has built that in. They're giving away a bunch of grants, hoping to show people that Anthropic is the best place if you're doing science. Okay, Base 44 is launching their very own AI model. And the reason why is, by the way, I didn't realize, but Base 44 is owned by Wix.
It's an incredible Vibe coding platform. I've been really impressed with it. It's not the one that I primarily use, because I'm just using Claude code for mostly everything I do. But if you're getting started, I think anywhere between Loveable and Base 44, those are my two favorite, Anywhere is also a great platform. But in any case, Base 44 is $100 million in annual recurring revenue, and they just launched their brand new model called Base 1 So this is their very own LLM. This is actually very similar to what Cursor is doing, by the way. But this LLM is trained on tens of millions of real user interactions. So what's interesting is because they are able to have tens of millions of users using them to create their tools. They can take all of those conversations, and I'm not sure if it's every single user or if they had people opt in. It's probably every single user, and it's just part of their terms of service. But they're taking everything that those people have been saying, and they're training their model to basically fine tune what people need when they're building websites, apps, and software. Now, what's interesting to me about this is, they know exactly, like when someone's building software, and you know, theoretically anthropic and codex, like from opening, I should have this as well, but they know every time that someone asks it to build something and it does it, and then they give a follow-up, they're like, oh shoot, every time someone asks for this, you know, for XYZ, they typically ask these next five things. And that could either mean that when they first ask, they don't know exactly what they want, or there's traditionally a bug that follows, or there's a gotcha, or something like that. But essentially, they can fine tune this model that they have to know exactly what they want to, what people are basically, the direction they're going before they even go there, right? They're like, hey, I want like a fitness app, and like, oh, these are probably the five features they want. So either they'll ask them right away, or they'll just start building stuff in, or they'll set up the architecture in a way that is ready to go. Now, there's a couple of cool things that they can do with these fine tuned models. One of them being that they can make them much more efficient, and so they're able to cut the costs for their users, which of course is fantastic. But also, if someone's talking to their model, and they're like, oh, you know what, our base one model would be a lot better than sending it to Opus 4.8, all of a sudden, they're able to switch people over to the more efficient model, save the user money, and they're also keeping exclusive data, right? Because if they're not sending data over to OpenAI or Gemini or Anthropic, all of a sudden, they're not able to train with all of that data. So we're also seeing the same thing coming out of cursor. So a bunch of different players that are interacting with code are basically intercepting the code and the conversations people are having and knowing what people want, and they're being able to train their own fine-tuned models with it. So Base 1, basically they're saying this is designed to replace Anthropics Opus for app generation workloads, and it's going to try to give them a really direct control over latency, cost, and inference speed. And they're not going to have to rely on any third-party APIs. And also, by the way, you know, this is a big cost for them, because when you get a subscription to Base 44, you pay them for tokens, and a huge chunk of your token bill just goes straight to Anthropic, right? It's like they're giving a huge chunk, but if they can do it directly on their servers, and if they have a cost-efficient way to do that, they're actually able to make more money. So Lovable that I've mentioned many times is, of course, their rival in Sweden, and they still use external LLMs. They have about $500 billion in AR, so they're about five times bigger. But they do have some really solid backing because Wix acquired them for about $80 million when they were only six months old with eight employees. Their parent company is now cutting about 20% of their workforce, so Wix is cutting down, but evidently Base 44 is accelerating. My prediction here is that if Base 44 really pulls this off, Base 44 will actually end up being much bigger than Wix, which is pretty wild considering they bought it for $80 million. It's already doing more than $100 million in annual recurring revenue, so they were just blasted past even the acquisition price. This was an amazing deal for Wix. All right, the next thing we ought to talk about is that X has just launched a posted MCP server.