Meet Moonshot, China's latest Al challenger

[00:02] Moonshot, saying its new model can compete with top offerings from OpenAI the latest headlines, we're calling in for tech support with Yahoo Finance's Dan Howie. All right, Dan. So, let's start there. China's moonshots uh with

[00:15] an AI model that says narrows the gap here with American companies. Let's just start there. Um what is Moonshot, Dan? Explain that. And then how significant is this news in your opinion? Yeah, Moonshot is a a AI development uh

[00:30] company AI AI developer similar to what you uh would see from you know anthropic open AI uh and it's uh open source uh or open weights rather uh so that means that people are able to download uh the software and run it if they can uh for

[00:46] free but a lot of people don't have massive server farms so what moonshot massive server farms so what moonshot does is they uh will charge people uh to servers or uh you know in some instances there are openw weight companies that

[00:59] will charge uh if you're going to use their uh application programming their uh application programming interface uh commercially uh but so this is uh Kim 3 uh K3 is what they uh they also call it uh and so it's

[01:14] consequential because of how uh powerful it is or or how near frontier it is. It's basically they say it has frontier level capabilities um just shy of level capabilities um just shy of anthropic stable 5 uh or openai's uh GPT

[01:30] anthropic stable 5 uh or openai's uh GPT 5.6 uh soul uh and those are their their top models right now that that are available. Obviously uh Anthropic also has mythos 5 but that's in limited availability for uh basically cyber

[01:42] availability for uh basically cyber security uh kind of measures. Um and so the fact that this company is able to to offer this uh shows that uh the uh Chinese uh AI companies are catching up with the the frontier US labs very

[01:56] with the the frontier US labs very quickly. Now uh it also uh however shows that people are increasingly uh using these models uh these open weights models uh because of how inexpensive they are to run. Uh and so that's one of

[02:09] the big differentiators when it comes to these types of companies and the the frontier companies that we regularly talk about the openis the anthropics the Googles uh and what have you uh they do have some uh open weight models uh but

[02:21] their big frontier models are are proprietary and so you have to pay to access them and so uh a lot of times the the cost is a lot higher to be able to actually access those versus something like a K3. I'm sure Dan, one question,

[02:35] as they listen to all this would be, okay, is all this historic American big tech AI capex, is that actually money well spent, Dan, if now we're seeing

[02:49] Chinese companies move in and create these strong models at much lower cost? they're they're absolutely going to have to answer. I don't think there there is an answer just yet. Uh, you know, this is kind of the not exactly but kind of

[03:06] the same discussion that we had with with DeepSeek when that came out. Um, I believe that was last year uh with their R1 model. Um, how they were able to train it up using uh you know low power uh GPUs or lower powered GPUs than than

[03:21] uh state-of-the-art um and provide a model that was very robust. Uh I I think this is it's it's not exactly the same, but it's it's one and the same, I think. You know, is the spending worth it? Do you actually need to be going all out uh

[03:36] with these kind of high cost uh uh moves, this big data center buildout where you're, you know, you're uh using this to train and then serve models up. I mean, that still seems like it's going to be something that needs to be uh uh

[03:49] isn't exactly easy. So, I think the data centers really uh are kind of maybe a if it's if it's easier to train and run these, then maybe they don't actually have to be as powerful as uh they currently are. But that that aside, um I

[04:05] I do, you know, think that the the anthropics and the open AIS probably themselves, well, gez, we're we're spending a lot of money to to train up these models to develop them. uh and then we have competition uh in China

[04:19] that is able to do it uh seemingly for for less money than than we are. Uh that was also part of the discussion with with Deep Seek uh you know at the time it was open AI obviously being the the main kind of company and the the kind of

[04:33] properly? Are they doing this right? So I think we really have to wait and see what happens now uh on you know the the flip side where where open AI and and know we had that report the other day uh from Bloomberg basically talking about

[04:48] how Google uh was delayed with its models. Google uh pushed back and said they're at and they're moving forward. So you know we we still have that that something coming out from anthropic and open AI. I think what we've kind of

[05:01] talked about generally is that these models are going to continue to kind of tick and talk back and forth where one is in the lead uh one is uh behind the again and now I think it's going to be more uh opened up uh globally as we

[05:15] forward and forward and closer into the frontier. about Dan. I mean just more broadly when you think about this great AI arms race, right? How quickly is that gap shrinking, Dan, between the US and

[05:32] >> It it has been shrinking and it continues to shrink. Uh, you know, we just saw uh Anthropic and OpenAI debut their their latest models, the Fable 5

[05:44] their their latest models, the Fable 5 uh and uh GBT 5.6 Soul uh last month. Um and so, you know, it's it's not as though it's been a long time uh since they came out with with this. uh K3 is a big step up from uh Moonshot's last AI

[05:59] model. Um so I think that's an important context there. Um you know obviously Anthropic and Open AAI have been leading in kind of the the frontier space and front frontier just meaning it's you know the the top of the top of the line.

[06:13] for for quite a while. So you know Moonshot uh hasn't uh and now they're they're kind of right up there. And this is all based on by the way on their evaluations. So you know I think third parties are going to dive in and and you

[06:28] know confirm that uh or provide their own kind of guidance on on what the capabilities of the particular model are but we are seeing this uh happen and you know I think if you look at um uh I think of you know Nvidia's uh Jensen

[06:41] models or openweight models uh during presentations and he always features uh the Himei models uh as well as the Quen models uh you know uh the the open- models uh you know uh the the open- source kind of uh uh area. Uh Nvidia has

[06:56] its own open- source models as well and you know those are are always you know you know those are are always you know relatively close to to you know the the competing models from the west and so you know at this point it it does feel

[07:08] as though that that gap is shrinking. >> On this same AI theme Dan Chris Mims at the journal out with a new column how AI's wider availability is good for China not great for open AI and anthropic. Mims asking if AI models turn

[07:26] out to be a generalpurpose technology like the automobile or electricity what can they uniquely offer how do you answer that Dan I I think uh it's going to come down to likely platform and ecosystem uh you

[07:41] know one of the things that uh a lot of these companies can do these these open source or open weight companies uh is offer surrounding services so that's you know you have an a piece of open source software uh you know the company that

[07:54] that puts it out is then able to provide a wrapper around it uh or you know lines and that could be something uh that that uh OpenAI and Anthropic and and their like uh lean further into we've seen them you know kind of do that

[08:09] sure you know we talk about uh their latest models but you know and they they powered their software but their software on its own is really I think what what's been driving uh their their kind of popularity If you look at claw

[08:23] code for instance uh yes that that does run on on anthropics claude uh software and yes that is a key element to it. Uh but they als my my point is that there's these different pieces of software that they could market that are run on their

[08:38] they could market that are run on their models that could be uh uh a kind of hedge against this kind of commodification uh of these uh model think that they have other ways around this. Um, and again, this is still

[08:53] pretty early for for us to, you know, just kind of throw our hands up and say, well, everyone is equal now. Uh, I don't think it's it's quite there yet. I think, you know, there's still also a lot of way to go in terms of, uh, AI's

[09:06] overall abilities uh, when it comes to these models. And so, you know, yes, uh, K3 is is very capable. Uh but I do think that you know obviously Anthropic and see more advancements from them as well as you know uh Moonshot Google Meta by

[09:22] the way they're working on their uh Muse uh line of models. So there's there's still I think a a good amount of room for this competition to continue. So is your point, Dan, is is this why you would see a Sam Alman moving into agents

[09:36] and software and hardware because he's building out this this entire platform, this ecosystem. >> Yeah, exactly. I mean, that's that's why, you know, I mean, obviously AI

[09:48] agents came along um you know, we were talking about them in 2025. uh that was blew up and then obviously this year and and you know uh early this year is kind of I think when it's really hit the the kind of broader zeitgeist uh but one of

[10:03] the uh things that they are doing as you point out is rolling out these these new pieces of software so they had uh codecs uh that allows uh you to do uh different types of uh programming work uh then they also have their own uh GBT work now

[10:19] that's a piece of software uh where you would do things like, you know, work through uh spreadsheets or, you know, put together uh slideshows, kind of the things that everybody on the planet despises having to do but does anyway.

[10:32] Um that's supposed to help you uh along those lines. Uh Anthropic has the same thing. They have uh Claude Co-work. Uh so, you know, you're seeing more of these companies lean into how they can serve up the models in different ways

[10:44] serve up the models in different ways that really provide, you know, a a boost obviously the user. And so the more they can sign uh up when it comes to different companies. Uh and so that'll

[10:57] over time. And so, as I said, the models are incredibly important to all of this, are incredibly important to all of this, but it's the software that they they kind of run under that is just as important.