Mere Morpheus

Mere Morpheus artwork

The Perpetual AI Machine | Staking MOR To Run Your Agent

49 minAudio

Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via their podcast app and to have them all in one place. If you'd like to participate they are held weekly on the @MorpheusAIs Twitter spaces. So enjoy Episode #67! Recorded: 2 April 2026 AI SUMMARY: This week Kyle, Bo and David discuss why "perpetual AI" on Morpheus now feels real. They unpack a new community calculator for estimating MOR staking or API credits needed to run specific models. They cover the first phase of Morpheus TEE attestation which proves what providers are running without trust and preview phase two extending verification to the GPU. On models they discuss Google's Gemma 4 Trinity and anticipation for GLM 5.1 plus tips on model routing and the API gateway's cost savings. Progress includes one-click agent deployment and Everclaw's new backup/export flow with a challenge to set up a perpetually running task. Timeline: (00:00:00) Intro (00:02:26) Measuring perpetual compute via community-built staking calculator (00:04:49) Two paths to Morpheus inference: marketplace node vs API gateway (00:07:05) Pricing in an open marketplace (00:10:40) Cost advantages and agent workloads: saving 80%+ with model routing (00:13:19) Right model for the task: routing strategies, tool calling, and Kimi vs GLM (00:17:50) Everclaw progress: backup, export, restore and bootstrap to local (00:25:09) TEE roadmap overview: phase 1 attestation live, phase 2 to GPU coming (00:29:43) Open‑source news: the Claude code leak saga and industry memes (00:32:34) Fresh models: Gemma 4, Trinity and rapid open‑source acceleration (00:35:28) Frontier cadence and GLM 5.1 expectations: near‑Opus performance (00:42:02) Home and SMB providers: the return of small‑scale compute with TEEs (00:46:30) Wrap‑up, community updates, and a challenge: launch one perpetual task If you would like to support me for the value I provide you can use the referral link below when depositing capital and we will both earn bonus MOR. Alternatively you could send me a token of appreciation directly to my wallet ;) Referral Link: https://dashboard.mor.org/capital?referrer=0xE935f231c99c04Ee0b4532a3d0BdA81B152a0384 Wallet Address: 0xE935f231c99c04Ee0b4532a3d0BdA81B152a0384 Connect with Mere Mortals: Website: https://www.meremortalspodcasts.com/ Discord: https://discord.gg/dX9BEhT3kz Twitter/X: https://twitter.com/meremortalspods Instagram: https://www.instagram.com/meremortalspodcasts/ TikTok: https://www.tiktok.com/@meremortalspodcasts Value 4 Value Support: Website: https://www.meremortalspodcasts.com/support Paypal: https://www.paypal.com/paypalme/meremortalspodcast

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Bowen BelmerAlright. Well, I don't wanna keep us too much dead air. It looks like we at least have Kyle promoted, David, myself. Let's at least kick off this week's discussion for everyone else who's already joining. So a few major things going on. Perpetual AI has arrived, and by that, we mean kind of what we have been talking about over the past couple weeks. But Morpheus, the protocol, is more usable for the everyday person today than it ever has been.

Thank you to the API gateway and all of the hard work from people who have contributed to that product, Dash subnet. But even more so than that this week, one of the biggest questions that I know I've seen in the public channels like the Discord and and have just seen other people engage with as well is how do I measure how much perpetual compute I have access to?

And so there is a community driven and built answer to that as well, which we'll be diving into this week, which is very exciting and very useful for anybody who's looking to stake more in order to have perpetual compute. Of course, one of the most important questions associated with that is, well, how much compute do I have and what can that compute actually grant me in terms of using various models? In addition to that, we do have some news with regard to open source models and expanding in terms of who's offering open source. I definitely wanna dive into

what those changes were, why we think they're happening, and kind of elaborate on that. And then lastly, and and if there's anything else that I'm missing, of course, anybody else, please feel free to jump in. Just a big congratulations to humanity in general for the rocket launch yesterday, which is very exciting Artemis two, which is not directly related to AI, obviously, but but it's just really cool to see technological leaps play out in front of our eyes. If there's one thing I've learned, it's that the Venn diagram of those interested

in space and going to space and those interested in developing AI and more specifically, DAI are pretty tight Venn diagrams. So it's very cool stuff. I would like to start out though on kinda what I was alluding to before, which is a new tool or or way to measure how much perpetual compute I'm actually getting when I'm staking to Morpheus. So I guess, Kyle, you're probably the best person to call on for that if you can provide an update to,

you know, what's going on there.

Kyle (Bow Tied Blue Fin)Yeah. Sure.

David JohnstonWait. What's going on with in respect to which aspect of it? I think he's talking about the the new calculator that Nomad Oh, all the Yeah. Shipped, which is which is awesome, by the way. I saw the, post on this thread. It looks, looks really cool. And, yeah, definitely, Burr. That's been the number one question I get. It's like, how many more do I need to stake?

Kyle (Bow Tied Blue Fin)Yeah. Absolutely. So the, yeah, the calculator is live at, I think the post followed here, but it's at tech.more.org/calc. You can also just get to it directly through tech.more.org and and click around there. The calculator yeah. I agree. It's it's super important. Right? We had a very basic calculator up a couple months ago that was pretty complicated to use, and you couldn't exactly get the information that you were looking for.

The new website that Alan, on the you know, one of the our key contributors, Alan, push live really kind of goes in at a deeper level. Right? What are you looking to do? How are you looking to do it? And how much is it gonna cost? Or or how much are you going to need to stake to get it? So it looks at both direct interaction with the marketplace. Right? You know, we we know there's two ways to access inference through Morpheus. One is by running a consumer node to access the,

decentralized marketplace. In that method, you're staking more directly, to create a session. So it looks at, alright, how long do I wanna create a session? Right? If I wanna create a a perpetual session, it needs to be, you know, one day long. For which model, how much am I gonna need to stake to that? How many how many concurrent prompts? Right? How many sessions am I going to actually need? And it helps run through through that mechanism. It connects directly with the blockchain to figure out, alright, what are what's the best bid right now for this model,

which models are active, and so on. So and as well as what's the current calculated price for how much I need to stake versus how long I can get. So really, really valuable there. The other side being the, inference API. So just looking at, you know, what are the the current prices, what are the models available, and then helping do some of those basic calculations of I'm running an agent, you know, how many tokens is it using, and then how much is that going to cost me, as well as the associated

staking through the API. I know I said a lot there on it, but I I actually I'm realizing there could be some confusion from what I said. There's two aspects to staking for inference on Morpheus. Right? One is if you're using the node and going directly through the marketplace. The second is if you're looking to just use the the Morpheus API and create an API key and stake for credits in that manner.

It's just a little bit different in the technical complexity. I would say larger users, who have more significant infrastructure would probably use the node directly to to kinda get a better deal, have a little bit more control, and have that end to end connection. Whereas users looking to migrate over from OpenAI or Anthropic or connected to an agent in a more simple and user friendly fashion that they're probably gonna use the API. And and, of course, they can purchase credits through USD, through Stripe,

or Coinbase Commerce, or they can stake to a builder subnet to get credit. So, sorry for for the the long explanation there, but I realized there could be some complexity as I was explaining. All good. No worries. Yeah. I think that's helpful, Kyle. And I would say that if you

Bowen Belmerdidn't understand it the first time, you're an API user. If you did understand it the first time, you're probably somebody who might go directly to the node.

David JohnstonYeah. Well and just to simplify it, go on the website, use the toggles, and it'll tell you exactly how many more you need to run your favorite model.

Kyle (Bow Tied Blue Fin)Yep. And exactly. But one last comment I would make there is let's all just remember that Morpheus is a two sided decentralized marketplace and the infrastructure for the marketplace. So depending what other providers come in or change, pricing is going to change. Right? We anticipate price is going to change favorably over time, but it's definitely going to be up and down based on who's coming in and providing at what capacity and and and, you know, what people are willing to pay. I assume we'll we'll adjust that as well from the provider side. So

I'll say right now, no one's allowed to say, hey. Pricing changed. Why? It's it's an open marketplace. Whatever people are providing, that's the price that they're providing at. And we invite more providers to come in and monetize some of their excess compute capacity that they have to to generate some more for themself some more emissions for themself.

Bowen BelmerYeah. Absolutely. And I I do have a quick question, Kyle. I'm looking at the site right now, and I see that on on the side of buying credits, which is probably where a lot of people will start. So I'm looking at the API gateway toggle for max convenience and then looking at buying credits as opposed to staking. Of course, my opinion is, like, the economically, the best model would be buying more natively and staking it. However, for somebody who's maybe using this for the first time, I see that it's measured in request per day in terms of where your dollar gets you, like, per model.

Is is that the request per day, is that, like, a a certain number of token output? Am I just missing that?

Kyle (Bow Tied Blue Fin)Yeah. Request per day is kind of an abstraction for simplicity. Right? Nothing is actually measured in request. They're measured in tokens from the API side. Requests are just, to help people say, like, how how should I think about this? How often am I using these models? But the real metric and the real, billing is done on the API on a token standpoint. On the and I'll take this opportunity to explain the actual marketplace again.

But when you're going direct with the marketplace, you're paying based on time. When you're going through the API, you're paying based on tokens. So think of one as I'm renting a GPU, so I'm paying for GPU time. The other is I'm paying someone who's hosting a GPU. It is just at a very high level. That's not actually how it is, but that that's a good proxy way to think about it. Okay. Awesome. And and it's so amazing. I mean, just seeing here the the cost savings, for example,

Bowen Belmerit is 81% cheaper to than running Claude's Sonnet 4.6 through Claude's API natively. Maybe that'll change with some of the news that we're gonna be talking about here shortly. But, yeah, I think if if more people even just purely look at this economically, it is, like, an incredible opportunity. I also really like how at the bottom here, we have the section for AI agents and automated workflows and estimating

what the cost would look like. So for example, daily research and reporting, like, that's something that I personally do with my agent that I've set up. And the fact that you could have it perpetually running for what is that? 400 more staked is amazing. I mean, like, that's, like, that's that's so crazy to think about compared to, one, what that would cost using a centralized provider, and two, just, like, the ability to do that technically, which six,

nine months ago, certainly a year ago, wasn't even possible. It's it's amazing.

Kyle (Bow Tied Blue Fin)Yeah. It it really is amazing. It really is. And and when you go back and look at the at the price differences, right, versus Opus or or versus Sonnet, I just want to remind everyone that most agents and and Everclaw has this in it. There is like, model routing is important. Right? So a lot of people are checking their emails with Opus, for example. Right? That's you you probably don't need, like, a multi PhD cluster to read your emails.

You can probably do that with, you know, a child. So let's make sure that when we think about kind of what models we're using for what actions, and we we use some intelligent routing there. Right? Like, it makes sense to get a 90% discount on on simple, repeatable, low intelligence tests. Right? I'm not totally against people using, you know, ultra expensive models for the most complex test. Right? Like, I'll still say that I use Claude Opus. Right? Maybe I'm a hypocrite, but I still use Claude Opus for,

you know, a couple things here and there when I'm coding, when I'm, like, at a a really difficult problem that I need to solve. And that's okay. Right? The like, we're going to see open source models come out this week that are gonna be comp comparable to the newest cloud opus, and and that's when when I expect I'll I'll switch over to those. You know, at the level of Sonnet, I have no reason to ever use Sonnet. There are open source models that are at Sonnet's level at such that discount that you said, 90%. So it's kind of I like to think about this as we don't need to go and and,

like, stand on a hill and and die saying, I'm I'm going to use this model for everything, or I'm going to never use a a closed source provider. I think that we're we're learning as we go, and the industry is changing so fast that, you know, we should still be flexible and use different things in different applications.

Bowen BelmerYeah. Absolutely. I mean, I know for myself, like, this is one of the things that I struggle with the most in terms of knowing what to do is is the agent selection or I'm sorry, is the model selection. Like, right now, I happen to be using Kimi k 2.5 for pretty much everything because I found it to be I don't even I mean, to be totally frank, I don't even know if it's optimized for, I mean, the basic things I'm using it for, I I think to your point before, like, most of these models could do pretty

pretty well. And so it's kinda just like, oh, where do I feel like I'm getting a good bang for buck? But maybe that would be a great addition to to this site or maybe another site that someone within the Morpheus community makes is like suggestions on actually, this would be a great agent. If someone could make an agent where I just describe to it what I would like to do and it makes a suggestion as to which models I will get the best bang for buck in terms of my staked more, that would be Yeah. That would be fantastic. Because, yeah, I there's

so many of Absolutely. Them

Kyle (Bow Tied Blue Fin)So I'll start on one thing. If you go to apidocs.mor.org and within the available model section, there's a little bit of guidance there on on good applications for different models. But further than that, I would say, Bo, you shouldn't need to know that. Right? Like, your agent, no matter what model it's on, it's going to be able to call web search functionality. It's going to be able to do research and pull blogs, and it should be able to it it is able to go and look at different benchmarks and understand your context and what you're doing and and execute that properly. One big example here is we all spoke about MiniMax two five and MiniMax two seven

and thinking about, is this great for agents? And it is great for agents in certain applications, but it either can't or or it doesn't do tool tool calling very well. So if you have the orchestration part of your agent that's running, it's not going to work well with with MiniMax two seven. But if you have it doing a specific analysis on, you know, something where the context is provided, then it's gonna work great. Kimmy, on the other hand, is really, really good at tool calling. So that's why it's probably working so well for you. So it's like

and and then I'll go one step further. Kimmy isn't good at super deep complex, you know, reasoning and analysis. So maybe that's when it would switch to GLM, for example. So it kinda goes right back to the point that, you know, you don't have to to, you know, take the the hammer to every everything you see, whether it's a screw, a nail, or or a staple. I mean, it it's putting the right tool for the job, I think, is where we're going to see some of the biggest improvements moving forward.

Bowen BelmerThat makes sense. Yeah. I'm wondering if, like I'm just trying to think this through because I I at like my level that I personally am currently at, it's to the point where like I'm, you know, myself setting up my own agent, so still making some of the decisions in terms of like what models I'm having it running. And I'm wondering actually, think, like, without a doubt, there's an opportunity here too for somebody within the Morpheus community to build just like a one click agent deployer, basically, where all of that stuff is kinda just like in a in a in a package for somebody to just access via

their device without making any of those decisions, if that makes sense. I mean, I know ultimately that's, like, what we're working towards. Right? Is people registering agents in the agent registry that other people can then use. Sorry. I'm probably talking about something that's, like, extremely obvious. I'm just I'm trying to think through, like, what is the lowest barrier to entry way for me to walk up to someone on the street and be like, you should try this. And, obviously, like, even just the API gateway plus this calculator is taking massive steps forward in that direction already.

But I'm trying to think through, like, what is the next version of that.

Kyle (Bow Tied Blue Fin)That's a great great point. David, why don't you

David Johnstontalk about that? Sure. You know, that's sort of what I've been driving towards with the tech underneath Everclaw is this idea that, you know, somebody can start, get going, and basically take all the friction out of the process. Right? I mean, that's one of the magical things about Morpheus is you don't need a clawed key. You don't have to know what a clawed key is. You know, you don't have to go poke around, you know, some settings or a terminal. You know, it's getting to the point where you're gonna show up to a website, and and I really like what you're describing, though, is I wanna be able to meet somebody,

tell them about Morpheus, and then literally go to a URL, and they click a button and that's their agent. And they give it a name. And, you know, if you if you notice, I just published a whole new feature on Everclaw that allows you to backup, export, and restore your agent to another device. So this this is a big deal. Right? So if you're running it on a Docker container or in a VM or somebody's providing a website,

they make their agent, then they can put it on their machine. They can export it, they can locally install it. So you can let people do the setup step first, get a taste for how cool it is, know, connect it to whatever they want, and then they can go through the friction of putting it on their machine or, you know, downloading it, to their computer, you know, and you're basically creating a bootstrap method. Right? Where now it just looks like any other website.

Right? And they don't have to understand how all of that works, in the back end, you know, and they can try something out, you know, know, they can play around with it. They can experiment, but then they can, you know, literally give one command and say, okay. Go ahead and install yourself on this computer. Right? Maybe they were waiting for their Mac Mini. Maybe they wanna use an old laptop, but they can go from that website experience onto their local machine with just a single voice or text command. Right? So that's where we're headed towards.

I'm actually developing that into a defined skill. So the the feature is actually built and shipped as part of 04/2002, the current version. But this will allow you to basically just call this this download my agent skill. And, yeah, then you'll be able to go from a container or web environment straight into a local machine. So, yeah, I think that that ought to be a big part of the focus is, like, how easy can we make this for as large an audience as possible to access?

Bowen BelmerYeah. Well, mean, that's I think what'll be the next chasm cross, if you will, is I'm picturing I'm picturing a site. It's like a landing page with one text box, so like a one password site where you don't even have a login. You just establish a password, and as long as it's a unique password, it just kicks you straight to your agent that is running on a virtual machine somewhere. And then to your point, if I elect to say, okay, I actually want to take this, like, in house and run it on my own hardware and or my own virtual machine, which we could also set up the optionality to do probably.

It's like a one click export. Now my agent doesn't live in this one password situation anymore. It now lives in, you know, the my own environment, basically. But, like, you could go on your phone and just in in one line of text, like, generate an agent that now, you know, messages them on Telegram or whatever. I don't know. I I see just like Oh, wow. You're you're giving everyone all your ideas. Well, I can't build any of this shit or stuff. Sorry. So I I

I'm fine giving away the ideas as long as someone builds it. Awesome. It's good for the ecosystem.

David JohnstonBut let let me just say it's getting built. It's getting built fast. And, yeah, it'll definitely be something to to cover in another one of these very soon because, yeah, I think that's exactly where where we're getting to.

Kyle (Bow Tied Blue Fin)Oh, well, that's amazing. And and we've seen the industry move you know, the industry is moving there. Right? We've seen, you know, we're probably going back a month now, but Cloudflare launched, you know, how to deploy your OpenGLAW agent in in, you know, three clicks here. I I think there's a lot of, tech probably being built right now around that. And and I'm all for it. Right? As you know, I think first is going to come

centralized, and then eventually, we'll get decentralized. Right? People, you know, Hetzner and and all these other, you know, VPS services, not necessarily decentralized, but once you start pairing many of them together and and create kind of a distributed network of them of, yeah, just go in and deploy this somewhere. Yeah. I think I think that's coming.

Bowen BelmerYeah. I I think you're right. I also just noticed you probably touched on this before, Kyle, but I was just messing around with the calculator again, and I just realized that and you can kind of, to some degree use one to to see the other. But if you toggle over to the running your own consumer node, it'll actually just show you how much more you need to have stick to run any given model twenty four seven. Yep. I didn't realize that was updating in real time when I was first looking at this, which is a really cool feature. And I feel like mean, I know it's not gonna be one to one, but you can sort of extrapolate

it to how it translates in the API gateway. Right? I mean, it'll be similar. It won't be exactly the same.

Kyle (Bow Tied Blue Fin)Yeah. It it's it's fairly similar. I mean, that that's kinda why we have the two tabs. The the conversions are a little bit different just based on some of the underlying technicals of of how the node runs versus how the the API runs on top of it. But, right now, you are there is essentially a a small fee built in for if you're staking on, the API, versus

Bowen Belmerstaking directly into the marketplace. But, you know, there are Well, but that's the benefit of the API. Yeah. You're paying for convenience, which is fine. I mean Yes. Yeah.

Kyle (Bow Tied Blue Fin)Yeah. And it's, again, we invite everyone to use the solution that best matches what they're looking to do. Right? This isn't some, you know, there there's no we're not trying to push people in either direction. It's do what works best for your application.

Bowen BelmerAnd if you have a better solution than some of the solutions we have, please tell us so that we can add it to the list of of things that we're promoting. Yeah. I I didn't until now, I didn't even look at the consumer node side because, again, like, it just is not something for my own use cases that I would ever use. But, yeah, that totally makes sense and 100100% makes sense. Okay. So one thing I wanted to dive into next, and this is kind of a segue from what we were talking about before.

Well, actually, I'm sorry. Before I do that, before I change gears, just wanted, one other thing on the tech..org site. So that that whole time we were talking about, if you go to tech.mora.org, there's kinda two pathways, choose your own adventure here. The the we were just talking about the calculator. There is also the TEE road map. I know we spent a lot of time talking about this last week, and this is certainly something we will talk about a lot into the future as it becomes, you know, more and more robust and relevant. However, Kyle, is there anything from the side of

the builders basically to to discuss on this this week?

Kyle (Bow Tied Blue Fin)Yeah. I'll just give a quick not a a recap because I I can't recap everything that we talked about last week, but I'll say that the TEE functionality that was launched through Morpheus is a let's call it a proof of concept, right, or or a an attestation of of functionality if if we wanna play into that, where phase one of what we're calling it is

you can now walk through the chain of what happened with my request from the origination of it, from your your computer or your VPS through the system into the provider. Right? And phase two I'll I'll talk more about this, but phase two, which we're we're working on, is being able to watch that chain all the way into the GPU and back. And, inherently, you can't get to phase two of being able to look at the entire process until you're able to look through the first half of it into the provider node. So it's kind of an an iteration,

and we just wanted to show everyone kind of that that middle area of I I have my consumer. I I see the provider, and and I have attestation that they are not, you know, doing anything out of the ordinary rate. They're not logging. They're they're not running any malicious code. We know exactly what they're running, and we can prove it, cryptographically. And and the we'll say trust me, bro, a lot is is something that that we don't trust you, bro. We we wanna prove it, and we wanna see it. So,

it it's really a a a major major, launch that we had with this phase one of it and being able to show this and something that doesn't exist in the industry right now.

Bowen BelmerYeah. Absolutely. And and, also, the site is extremely well put together and easy to follow. So I do urge anybody, who is interested in how this works and how this is being accomplished to check out the tech.mora.org/tee, or just tech.mora.org and click kind of that that choose your own adventure path pathway through the links. And this is I we'll we'll talk a ton more about this over over the coming weeks and months, and I suspect that this will end up being one of the largest

I don't know if moot is the right word. Like, competitive advantages that Morpheus, the protocol, has to to relative to other things in the DAI space. Obviously, there's a lot of people building in the DAI space. Morpheus has had the same guiding, like, first principles the entire time, and so it's no surprise to me that we collectively, as a community, are are, like, among the first, if not the first to ship, you know, verifiably with physics

private compute in this way. So it is very cool, very much excited for the no need for trust me, bro, anything going forward. And just, you know, as a reminder to everyone listening, that's not trustless systems and permissionless. Trustless systems are not built as an indictment against any individual or any group of individuals, but rather just to remove the need to have to do it. You know, it's not like, I don't want to trust you. It's that why not just have a system where trust is not a requirement?

Because we can just prove it. You know? There's it it just completely removes that element from the equation, which is really awesome, and and extremely exciting. I did wanna transition, though, and sorry. I kinda took us back. I just wanted to make sure we touched on that as well since it's tied to the same site. But, David, you touched on this when we first jumped on the call here that there is some pretty big news in open source that being,

David Johnstonyou know, Claude had a had a pretty big announcement. Can you give a a brief overview of that? Well, it's it's sort of interesting. It's going back and forth the last few days. But for everybody that saw, Quad may or may not have released accidentally a bunch of their source code for Quad Code, and people sort of noticed that it was out there and real quickly refactored, mirrored, and copied something like 500,000 lines of code.

And so, you know, lot of sort of what they were planning and and things on their roadmap got leaked. One of which was was a sort of an Easter egg or a part of their their April fools strategy for this year. But it was funny yesterday, they put out a an announcement that nothing had been leaked and that this was an elaborate plan by, you know, their their engineers to, you know, provide, you know, an example of what a leak would look like. But I think the announcement itself was the April fools joke. I think they're sort of acknowledging

that it happened. It's kind of embarrassing, but, you know, it just happened to coincide with, like, the day before April fools or whatever. So I'm pretty sure the leak was real. I haven't I haven't gone in and kind of seen all the things people have, you know, sort of garnered from the knowledge. But one of the funny memes that came out of it is Quad is now more open source than OpenAI. And so that's a side effect of this leak is that's what you do in open sources is, you know, sure show how you're building and what you're building and and things like that. So, anyway, interesting set

of events and so, you know, we'll see what happens, but it sort of speaks to the fact that, you know, agents aren't perfect and they can make mistakes like include, you know, references and things you didn't expect and, you know, it was it was probably somebody moving quickly and and shipping code and, you know, just mess missed that one test or, you know, thing to check if if something was exposed. And so, anyway, kind of a a funny circumstance, but I I don't know what you've seen on that boat, but the the memes are pretty pretty entertaining.

Bowen BelmerYeah. Was gonna say that was pretty much it, and mostly a lot of stuff making fun of OpenAI as well, which is I'm always up for. I'm pretty sure my personal journal is more open source than OpenAI is at this point. But I digress. I I believe there was additional news. Were there some other models open sourced this morning as well? Thought I had seen that. So a couple of big things came out.

David JohnstonOlama, just literally an hour ago, was highlighting the fact that Google DeepMind has released Gemma four. And so this is something you can now run locally on Olama, And they put out different size models, you know, 4,000,000,000, 2,000,000,000, 26,000,000,000 parameter versions and even up to a 31,000,000,000 parameter. But if you look at the benchmarks, it's pretty amazing. These are relatively small models, and they're benchmarking at where GLM five is right now.

And that's impressive considering GLM five is what 700,000,000,000 parameters. It's a much much larger model and requires much stronger infrastructure to run. Right? So this is just this this week, today, this hour, another acceleration for open source models that one, this is open source from an American company. Right? Because there's sort of a growing, you know, criticism that, like, oh oh, the only open source stuff is coming out of China. That that's not true. Google's putting out some good models.

Mistral can continues to put out good models. Right? I just saw another one, Trinity. I haven't had a chance to check it out yet, but also another American open source model that's supposed to be benchmarking near the frontier as well. So anyway, really encouraging to see because what this means is you can get that GLM five type experience now for 20x less hardware. Right? And this is sort of what we continue to see and what we saw last year is the cost of tokens

and and the quality of those tokens, you know, became a 100 x better last year. And and I think we're on track to see effectively the same thing this year. We we've almost become normalized to the fact that whatever model came out a few weeks ago is gonna be 10 times smaller and 10 times better, you know, and cheaper, you know, in short order after it's released. But great to see, and credit to, to Google for putting the Gemma four, series out there today.

Bowen BelmerYeah. That's been a little bit my takeaway as well as I feel like, it's it's kinda like when streaming got really popular for, like, home TV and movies and that type of thing, the amount of people who would be like, I'll just wait till it, you know, goes on streaming instead of going and seeing it in the theater. I feel like I've now reached that with Frontier models. It's like, oh, I'll just wait for it to go, you know, 10 x cheaper in a handful of weeks or months when the next frontier model comes out, and then I can try it.

And then it's like this weird thing that we've adopted and gotten used to over two years, you know, which is crazy. I'm sure it'll just continue to accelerate in terms of timeline. Sure. I saw something that was it, like, every 23,

David Johnstonthere's a new frontier model or something like this right now? Well and and the one everybody is hyping and expecting right now is GLM 5.1. So they've already released an API where you can try it out. Doesn't really fit my privacy settings, so I'll skip that stage. But they expect to release the open weights for 5.1 in a week or two. And if you look at the benchmarks, it's 96% of Opus 4.6. Right? Which is just incredible,

as far as performance. So if even if it's it's a larger model like five point zero was for GLM, you know, that's that's still gonna be a huge moment for open source. Because as Kyle said earlier, like, there's still complex coding questions where you basically have to go back to Claude because there's not an open source model that can sort of get it done. Right? And I I think people were were just about across the chasm where we'll finally have a model equivalent to Opus 4.6

that can really have that kind of accuracy with coding and agentic tool calling that you need to do those really tough tasks. So I think that's gonna be a big deal. I can't wait to see the open weights get released. The folks at at Venice have already said, hey. We're gonna be listing this immediately, which means you'll see providers on Morpheus listing it pretty quickly after it comes out. So I think that's a moment that we should really lean into as a community because as soon as that happens,

the easy explanation that you can use Morpheus to save 90% on your Claude bill becomes even more robust. It becomes even more believable if there's a 5.1 model that really basically equates to Quad 4.6. Right? And so then the cost savings is just too compelling to ignore.

Bowen BelmerThat makes sense. Well, I expect, like, to your point, we will we will see the latest and greatest on Morpheus rails pretty much as soon as you can access it anywhere. So that is that is very exciting. Yep. I'm trying to think if I had any other DAI news. I mean, at least not to my knowledge anything specific out of Morpheus or the Morpheus community. Did you guys have anything else? I mean, otherwise, obviously, I just briefly touched on it. The other, like, huge news in tech right now is the,

you know, mission around the moon. And I suspect that as time goes on and we send more rockets to space, we will have more and more AI integrations into this type of thing. And privacy, perpetual access to compute, and all of these things will remain further and further relevant both literally and figuratively.

David JohnstonYeah. I mean, it's really been cool to see, and I think it's inspiring for a lot of us that are in in the space since we were kids. Though I I do look forward to the next launch. Hopefully, that's a a Starship launch. The current Artemis hardware was basically designed in the nineteen seventies. And so I'm glad I got off the pad. I hope they have a safe mission, you know, Godspeed. But, you know, it it will be sort of amazing to see it go from, you know, their little five person, four person vehicle

to something capable of carrying a 100 humans, to the Moon on on the next launch. So NASA's contracted with with SpaceX to do the actual landing. Right? And and sort of if you've been following the inside baseball, the new NASA administrator has been canceling sort of like all the random side projects that accrued over the decades as they tried to keep the whole Artemis program alive, like the gateway. Why is there a gateway? Why do we need to park in orbit before we land? There's no really good reason, so they just canceled that.

And the hardware that they've got today is all expendable. So the Orion capsule, the whole booster and everything, it just falls into the ocean. So I think the big unlock is as soon as it moves to reusable spaceships, it's just going to be an incredible next level. Some of the Falcon Nine's have been reused 30 plus times, so I think we're going to kind of look back and laugh like, What do you mean you used to throw the rockets in the ocean? Like, would you throw a plane in the ocean after you used it once? Like, that's that's crazy. So yeah. No. It's it's really good to see. Like, all tech seems to be accelerating

thanks to all of these advancements. Right? There's a almost like a a cool convergence, really happening.

Bowen BelmerYeah. Well, I'll say that the the, media coverage from SpaceX was certainly missed on this one. It made me appreciate it made me appreciate so much more, everything that they were doing over there. I actually have a camera instead of, like, a little icon that somebody made in paint, you know, showing the document. Great example. We're still on Trust Me Bro with NASA. When are we gonna move off off Trust Me Bro? I need live streams that are recorded on chain so I know that this is real.

David JohnstonYep. No. It's it's it's really cool. Well, you know, one more thing on the Morpheus side I want to to highlight. You know, we talked about earlier Morpheus is a marketplace. And one thing that I see Scott and others developers are focusing on is getting to a one command install for the Morpheus inference provider. And I think that's going to be another big unlock because as all this demand shows up where people want to use

Morpheus inference, we're going to need more and more providers to step up. And there's plenty of people that want to provide GPUs and it's been some of like solving this catch 22 right, where you need enough demand to attract the supply. Okay. Great. The demand is showing up, and it's it's starting to scale. So just wanna point that out. Maybe we can highlight that on the call next week. But, you know, I think having, you know, easy onboarding for for demand for people that wanna use it, but also easy onboarding,

for people that wanna provide, inference. I think it's gonna be really important.

Bowen BelmerYeah. Absolutely. Totally agreed. One, I I think that that'll end up being the renaissance of, you know, what was the so prior to, like, Ethereum going to proof of stake, I feel like at home mining was such a common thing. You know, like, every single and now I feel like, for example, Bitcoin mining is so institutionalized that there's very few people who are, like, running mining farms at their home. But I do think we will get to a point where people are providing compute

force for either Morpheus literally or something like Morpheus on their own because they won't necessarily be hosting, like, these large models that everyone's used to using. I suspect that there will be more specialized, like, fine tuned models that people will be choosing to host on their own. So, like, I envision a future where you have people who are designing and trading these highly specialized models for very specific use cases and, like, certain types of agents will access them. And then they'll be able to host them in their home or their home business or, like, small business

and run them in the trusted execution environment that we were talking about before. So even though it's technically, like, a central provider who's giving you access to that model, we can at least verify that it's still private. And that, like, completely opens the door, I think, for what's possible.

David JohnstonYep. 100%. Especially as, you know, Gemma four and we're seeing all these powerful models continue to shrink down to, I mean, 32,000,000,000 parameters. That's like 24 gigs. Like most people can run that on on their laptop. Now you still need good opt up time, you still need a good Internet connection. You know, there's more to being a provider than than just having a laptop, and I think it'll continue to be people, let's say, that want to really take it seriously

and always have the hardware available and and continue to improve it, but the barrier keeps falling. To your point, it's not you need a 100 megawatts of power in a giant data center. You could do this at a relatively small level and be a good provider if you spend the time to truly make it a professional setup. And I think we'll go through that same evolution. Bitcoin mining started with a bunch of amateurs, and laptops.

They maybe didn't mine a lot of Bitcoin, especially after the GPUs landed, but the people that got serious about it and treated it as work that they were, you know, excited about, you know, professionalizing it went right up that scale. And I think we'll we'll see the same thing here.

Bowen BelmerYeah. Absolutely. The other thing I look forward to in the future is as the, like, agent registries get built out too, I I look forward to a time, hopefully, the not so distant future, where every week we're able to check-in and basically be like, hey, did you see how, like, the popularity of x y z agent moved or, like, x y z agent jumped a b c agent in being the most commonly used for this use case? And, like, a a competition will ultimately arise

David Johnstonin that capacity as well. So Cool. And that's that's a good point. Like, KPIs, I think it would be really cool on these weekly calls to, you know, pull up some of these great visuals that Thomas or Alan and others have made and start tracking, you know, the usage and seeing, you know, how many tokens are are being consumed and what is right now, you know, the most popular model. Like, there's some great visuals now on these on these websites that they've built. They kinda go through that and just like, wow. Like, up into the right, like, we're seeing x amount of growth. Because I think that tells the story

that open source continues to win, and I think ultimately will win. Right? We've already seen overall use of open source go from 14% last year to 31% this year in just like the last six to nine months. Right? So clearly open source is on track to be a majority of AI in the future, and and Morpheus has been pioneering that for a while. So totally, good thing to to check-in on.

Bowen BelmerYeah. Absolutely. I will make a note of, note of that to try to flag for next week's call and subsequent calls going forward to see where that's at relative to weeks prior.

David JohnstonAwesome. Absolutely, man.

Bowen BelmerVery cool. Well, guys, that was all I had for this week. However, I would like to open it up if there's anybody anybody in the audience right now who wants to pop up if they have any questions, inquiries, things to report from their side, or if or if you, Kyle and David, have anything else you wanna make sure we cover?

David JohnstonI think we're good from my side. You know? 5.1 is coming. Trinity model just dropped. Gemma four. So lots of stuff to try out. I just posted the new version of Everclaw, so check that out. It's got a bunch of security upgrades. It's got this backup and export function built in. So, you know, I I also wanna thank, you know, I think it was Ether dash BTC, the developer that gave me a bunch of issues that I was able to turn into code

over the weekend. So very much appreciated. And to Alan and everybody else that's been giving me feedback and next features to build, it's it's definitely getting more and more robust every day. And, yeah, let's let's double tap next week, Bo, on on what you're talking about the web interface. I think that's the next big unlock.

Bowen BelmerSure. Absolutely. Well, in the meantime, my call to action for anyone listening to this is to, once again, set up your agent, give it a try, use Everclaw, establish perpetual compute for at least one task in the next week. And what I mean by that is identify one thing that your agent can do for you, figure out essentially how much more you would have to have staked or what to what charge you would have to go to on the API gateway

in order to do that perpetually, and have, like, at least one perpetually executed task. That would be, like, my challenge to everybody listening to this right now, and we will all together take these what what are, I guess, baby steps today, but are really, like, massively forward in in technology and, like, the way in which we use it. And, yeah, that's that's kind of my challenge everybody is to is to start doing those things which over time

will eventually equate to this, like, very personal super intelligence that you yourself will have access to that works, you know, in your own best interest, which is the future that we're often talking about here. And it's these small steps that will add up over time to be, like, the huge, large spanning outcome. Awesome. Well, thank you, guys.

David JohnstonThanks, Bo. Talk to you guys soon. Keep it awesome, everybody.

Open the transcript file

Hosts and Guests

Hosts

Guests

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Video Is Coming! | Morpheus Is About To Get Far More Entertaining

    52 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  2. Local Models, Foreign Affairs | The Case For Private AI Archives

    53 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  3. Let's Talk Node NEO | Step By Step Progress

    51 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  4. Hello From Hermes! | A New Way To Onboard To Morpheus

    43 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  5. What Agent To Agent Connection Actually Looks Like | Inside The Morpheus Agent Stack

    58 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode