Everyday AI Podcast – An AI and ChatGPT Podcast - EP 584: ChatGPT's New Open Source Model gpt-oss: What it means, the risks, and more

Episode Date: August 7, 2025

This ChatGPT announcement is more important than GPT-5.Seriously.This week, OpenAI (kinda) quietly released its first open-source model since 2019.Us AI dorks are talking about it… but the business landscape is crickets.(As everyone gets hyped for GPT-5 today.)But…. Hot take on a Thursday shorties: ChatGPT’s new Open Source Model will be a bigger step forward for AI tech than GPT-5 and it’s not even close.Join us to find out why, how business development could change, and who will be the winners and losers.Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Thoughts on this? Join the convo and connect with other AI leaders on LinkedIn.Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:OpenAI Releases GPT OSS Open Source ModelComparison: GPT OSS vs GPT-4 Level ReasoningImpact on AI Industry Competitors & StrategyApache 2.0 License vs Meta Llama RestrictionsBusiness Benefits: Local, Secure, Free AI DeploymentTechnical Specs: 20B and 120B Parameter VersionsAI Model Customization, Fine-Tuning, and Edge UseWinners and Losers: Nvidia, Google, API ProvidersEdge Computing and On-Device AI FutureOpen Source AI Risks and Safety ConcernsGlobal AI Race: US vs China Open SourceAcceleration of AI Innovation and Model DevelopmentTimestamps:00:00 "ChatGPT's Game-Changing Open Source"05:27 Open Source AI Models Explained07:23 OpenAI's New Open-Source Model11:30 Affordable High-Performance Language Models15:54 Meta's Shift Toward Proprietary Models17:56 "AI Model Customization and Deployment"20:38 Leveraging AI for Cost Efficiency26:13 OpenAI's Strategic Competitive Advantage27:58 OpenAI's Strategic Dominance Forecast31:33 "Anticipating Google's Gemma 4 Impact"35:18 Apple's Future in AI-Powered Phones39:11 AGI: The New Global SuperpowerKeywords:GPT OSS, OpenAI, ChatGPT open source, GPT-OSS, GPT4O level reasoning, Open source AI model, Apache 2.0 license, Reasoning model, Local AI models, AI edge computing, On-device AI, Downloadable AI model, 21B parameter model, 120B parameter model, AI model fine tuning, Commercial use AI, Chain of thought, Agentic tasks, Tool use AI, Secure AI deploymenSend Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI Show, the Everyday Podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the All In One Creative AI Studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. Everyone, rightfully so, has their eyes on OpenAI's GPT5, which is going to be released in hours.

Or if you're listening to this podcast a little later, it was just released. But there was actually a different Open AI release this week that I think is actually bigger than GPT5. hardly anyone aside from us AI dorks talked about it. And I think it's actually going to be the pivot point in the AI race when we look back many years from now. So that's why today we're to be talking about Chad GPT's new open source model, GPTOSS. We're going to talk about what it means, the risks and who the winners and losers are. I'm excited for this one.

We're going to be uncovering and unpacking a ton. in a very short amount of time. All right, let's get into it. What's going on, y'all? My name's Jordan Wilson and welcome to Everyday AI. This is your daily live stream podcast and free daily newsletters helping everyday business leaders like you and me. Not just keep up with all this news, but how we can use it all and leverage it to get ahead

to grow our companies and our careers. Starts here on the podcast, unedited, unscripted. But where you actually go and be the smartest person in AI, that's on our website, your EverydayAI.com. Go sign up for the free daily newsletter. We're going to be recapping. not just what's happening in the world of AI news, but the most important takeaways from today's show

or maybe some stuff that we couldn't get to, as well as you can go listen to almost 600 episodes now for free on our website. And if you do want the video version ever, you can always go find that on our website, click the episodes tab. And if you want the AI news, like I said, go check that out in the newsletter, but let's talk open source. Great name from OpenAI here, GPTOSS,

as if the thing to help the news, the alphabet soup of model naming was to throw more alphabet letters into the soup. Anyways, this is pretty big. Open AI just released a free open source model. And it is on par with the model that most of us were using nine months ago, which is GPT40. It's on par with it. And it has reasoning. So let's talk about this new.

models and there's actually two of them. So this is their first open source model since GPT2 in 2019 before many of us were even using this technology many years before chat GPT came out. Also

it does have a lot of higher tier capabilities that you wouldn't necessarily think would be in an open source model, such as reasoning and it's trained on techniques from OpenAI's Frontier O3 model. And this is a direct response to the Chinese open source movement.

Whether open AI is going to admit that or not, that's the truth. As the AI labs here in the U.S. were taking the proprietary approach aside from Mata, the Chinese AI firms went all open source, and that's really forced open AI's hand here. And it is available for free for anyone today to download and use. that is why I am saying this is going to be a pivotal point when we look forward. Yes, GPT that's being released today. It's going to get all the headlines, right?

But it's just a upgrade in a model. This is a completely different direction that the business world is now going to travel in. That's not hyperbole. I'm going to tell you why. And this, like I said, it does completely. change the trajectory of both open and closed source model development. There's no going back now. This thing is out in the wild. People are going to download it. They're going to fork it. They're going to distill it. It's got to get wild on the AI development front. And this is going to

force all the proprietary AI companies, the big AI labs, to shift their strategy as well. They're either going to have to release better models faster or they're going to have to cut prices significantly. This is really going to cut out, I think, the mid-tier market and kind of the one-C kind of tier of model makers as well. So on today's show, here's what we're going to talk about. We're going to go over how Open AIs first open model since 2019 ended their proprietary-only strategy and how their internal business model has shifted, why businesses can now download GPT-4-level reasoning and run it locally without APIs. We're going to talk about the winners and losers.

One of those winners is Ambidia, while API only providers could face an existential threat. And we're going to talk about why powerful AI becomes uncontrollable once it's released globally and why I think this will be bigger than GPT5. All right, let's get into it. So here's open source. If you're confused, maybe you're just, you just go to chatGBT.com and you're like, all right, I don't really understand this whole API,

open source thing, why is it a big deal? So right now, there's two different versions of this model. There's a 21 billion parameter model. And then there's 120 billion parameter model as well. They're completely free. All right. So normally if you're using chat GPT, you're paying, you know, on the front end, you're paying $20 or $200 a month for the pro or the plus plan or you're on a free plan.

But when you're on a free plan, obviously the companies are training on your data, right? in training on how you interact with the model. This is different, right? With an open source model, you download it, you can cut off the internet and use it. It is local. It is secure. It's relatively fast as long as you have a souped up computer, right? You do need a fairly powerful computer to run the 21 billion parameter model.

You do need at least 16 gigabytes of RAM and a newer GPU chip as well. So here's some other things that I think are going to change, right? Because this is, you can download, own this and control it essentially forever. And we're going to talk about the open source licensing that Open AI went with as well, which I think is pretty interesting, especially if you're sitting in meta's seat. So now I think small companies and startups are going to be able to gain access to high level AI. without restrictions, right, and without having to raise capital or without having to run up, you know, tens of thousands of dollars in API costs, which a lot of companies do.

You can run this now completely offline with configurable reasoning as well. That's great. You can run it on low, medium, or high reasoning. So this is a reasoning model, an open source reasoning model. Crazy to say this out loud from OpenAI, right, which all of their critics have been calling them closed AI for many years. You know, the CEO, Sam Altman said that they maybe took the wrong approach on the open here, but it rectified that fairly quickly here.

And so you can modify, fine tune, and integrate this open model into products without even asking open AI permission or without paying a dime without having any ongoing costs, right? If you have the hardware, you can run this locally. It doesn't require a ton of skills either. And I'm going to tell you the different ways that you can download it and run it today. So like I said, two different versions, the larger version. It is, according to Open AI, this is their large open model designed to run in data centers

and on high-end desktops and laptops. I believe, as an example, the highest souped up MacBook Pro, M4, I think it's like 64 gigs, which is like how do you even fit that much RAM in a laptop? That can run the 120 billion parameter model. And then the GBTOSS 20B described by OpenAI as a medium-sized open model that can run on most desktops and laptops. So yeah, you do have to have a newer laptop with at least 16 gigabytes of RAM. And I do think in the future, you know, an iPhone will be able to run this.

Not today's iPhone. maybe the iPhone 17 will be able to run this version or if OpenAI updates it in the future. And that's, I think, one of the big plays here that we're going to get to later. So let's talk about some of the capabilities. So again, live stream audience, you can see it on my screen, podcast audience. I'm just reading these couple of parts straight from Open AI's website. So kind of the four big capabilities, that they are talking about here is the permissive license.

It's designed for agentic tasks. It's deeply customizable in the full chain of thought. So on the license side, the models are Apache 2.0. So you can build freely without worrying about copy, copy left restrictions or patent risk, whether you're experimenting, customizing, or deploying commercially. On the agentic side, you can leverage powerful instruction following in tool use. That's the other thing. Yes, there is tool use, Python, website.

search, et cetera, when you are using this on your local machine. And then you can also follow along with the chain of thought. It's deeply customizable. So you can adjust the reasoning effort to low, medium, or high. Plus, you can customize the models to adapt to your use case with full parameter fine tuning. So you can literally download this, fine tune it, run it locally, run it on-prem, run it securely, run it without Wi-Fi, right?

And then full chain of thought. That's the other thing. This is a big, a big jump ahead in capabilities when you're talking about a reasoning model, right? Kind of quote unquote old school transformer models, right? They're essentially very advanced auto-complete models. Reasoning models are a lot different that you can kind of see their chain of thought. So these are models that kind of take time to think and they plan ahead like a human would. And obviously the results that you get from these reasoning models are much.

better than you get from transformer models almost in every single case. But then you can see the chain of thought. You can see, right, which is so important if you're a company using an open source model, you need to be able to kind of audit how it gets from, you know, question A to answer A, which is pretty important there. Benchmarks pretty good. We're not going to go through them all. We'll leave a link to them in the newsletter. One, I like to talk about even though it's kind of an older benchmark by now, MMLU. There's a lot newer ones, including like MMLU Pro, but this is essentially, I'd say it's like an old school ACT for large language models, right, but they're not trained on the data set. So the 120B version got a 90 on the MMLU,

which is crazy. The GPT 20B version, so the small version that can in theory run on a phone, not on an iPhone, but it can run on other phones technically locally. Y'all, this got an 85.3 MMLU. So that's essentially the same score as GPT40. Let's think about that. You have a model as capable as GPT40 that you can go download, fine tune. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI Assistant now live in the Adobe Firefly app, the all-in-one creative AI studio. Powered by Adobe's creative agent, Firefly AI assistant lets you start with your vision,

just describe what you want, and shape the outcome as it takes form with the assistant. The assistant orchestrates multi-step workflows, drawing on 60 plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time.

You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adopi.com. Use commercially without paying a single penny. You're not running API costs, right? And let me explain that for our non-technical audience, right? Many of us, we go to chatchipt.com, pay, don't pay, right? But you're using a front-end chat bot.

Then you can use it on the back end. So maybe your business builds on top of this technology. And you're essentially paying a certain price for every million tokens, right? So you're paying your usage. And sometimes that's a couple of hundred dollars a month. If you're a smaller business, if you're a bigger business, you know, could be hundreds of thousands of dollars, right? But you're paying for usage on the back end and maybe,

you have hundreds or thousands of customers that are actually using open AI's technology through the lens or through the API kind of pipeline that your company is connecting it with your data. This is free. This is nothing. You're paying nothing. That's why open source is really powerful and it does change the game. So let's even talk about the open source itself because it's different, right?

When we're talking about large language models, we don't talk open source. open source a lot because for the most part, companies aren't playing there. Because the monetization strategy on open source, as you can imagine, it's hard. How do you make money when you give everything away for free? I think opening eyes going the scorched earth approach, but we'll talk about that here in a minute. It just so happens there, released the GPTOSS two days before they're releasing GPT5, which will not be free. Right. So my thought is they're just trying to.

to make every other model obsolete or essentially obsolete and just force people, hey, if you want the best of the best, you go to GPT5 and we're going to knock out mid-tier competitors almost completely by putting a model as good as almost any of them out there. Right. And the top, you know, I'm guessing it's going to be the top five, top three percent of models available worldwide, maybe even better than that. We'll see once all the third party benchmark. come in, it could be a top two, top 1% model.

Why are companies going to pay? Right. Anyways, let's talk about the open source license and it's much different than Metas. All right. So opening I is using the Apache 2.0, a very common open source software license. So this essentially permits unlimited commercial use without user caps. There's no geographical restrictions.

There's no fees. It is truly free and open. Meta's Lama, a lot of people don't understand. It's not like that. So it blocks the biggest tech companies from commercializing on it. There's geographical restrictions as an example. Right now, you can't use multimodal capabilities of Meta's Lama in the EU. And there's a lot of other kind of fine-tuned restrictions on Meta's Lama.

So a lot of people, especially in the open source community, say, Lama is not technically open source, you know, Meta is trying to essentially define open source, which I guess they can do that. They're one of the biggest companies ever in the history of technology to get behind the open source technology the way they have. So essentially, Meta's trying to redefine open source technology. But here you go. Open AI just came in with way better licensing. There's hardly no restrictions.

Also, Apache 2.0 includes patent protections, making enterprise adoption legally safer, right? which is something that Apache 2.0 provides and Meta Lama is not on there. So right away, you can see how this is not going to be good for meta. And we've also seen reports and rumors over the last couple of weeks that meta's new MSL team, their meta super intelligence team, is thinking about maybe going to proprietary models and going to paid models or making certain models paid. right so the open and free uh kind of meta might be starting to go down the paid route and then the open a i which has been dragged through the mud and saying they're only closed and proprietary why are you

called open a i yeah now they're taking the opposite route and i do want to highlight this a little bit more about how big this apache 2.0 open source licenses so like i said that is unlimited commercial use there are literally thousands of amazing businesses that are now possible that just weren't possible, right? There was no open source reasoning models that at least people here in the U.S. could safely use and safely build upon. This is it, right? This is the gold rush, right?

because I think a lot of times when we talk about AI development, if you wanted top tier model, if you wanted that top two, top three percent model in the world, you could have the best business idea in the world. If it takes off, you're still paying for it. This is different.

This is free. This opens up so many, so many new business opportunities. If you are an entrepreneur, you have to understand what's on the table here. There's no user limits. There's no Revenue caps, there's no licensing fees.

It's literally, truly open source. The only requirement is you have to include OpenAI's copyright notice in your code files. You don't have to attribute anything else to them. And then, like I said, with the model itself, you can modify, fine tune and sell products without even asking permission or applying for any program. So as an example, if you're a huge bank, you can deploy this for customer service. You can fine tune it according to your needs. Never have to spend a penny. right so maybe you've racked up a monthly six figure bill building something on the API side well

you're saving money now so I want to talk about three ways you can actually use this because you're probably thinking okay how can I use this I have a computer with 16 gigs of RAM right I haven't knew enough computer how do I do this right there's a lot of different ways but I'm to talk about well technically four but three and a half different ways right so one is local tools. You can download certain programs, right? So I do this on airplanes, right? Sometimes I got to get a newer laptop with a little bit more power. But, you know, I have some of the meta models that, hey, in a crunch, if I'm on an airplane, no Wi-Fi, I can open up different local tools that you can download. So you have, as an example, a Lama, LM Studio,

things like that you can download. So essentially, they're software programs. You can download them and save them to your computer. And then at the same time, you download the model itself, either the 20B or the 120B. You can download that from HuggingFace, which is the leading open platform to download models in the world. Okay. And then that's it. You connect the two. You're good to go.

You can literally turn off the internet. All right. Then there's cloud platforms as well. So obviously, AWS from Amazon, their bedrock platform, Microsoft's Azure, hugging face, fireworks, together AI, etc. But then you can also do a hybrid strategy for your company. So you can do cloud training with edge deployment for regulated data compliance requirements. If you need certain things to be processed on-prem, you can do that as well.

So you can do a hybrid strategy. Or if you're just like, all right, well, maybe you don't have that powerful of a computer or maybe if you just want to play around with it a little bit to see if this is for you or for your company. I love that open AI made this. they made a playground. So you can go check it out right now. We'll have this in the newsletter as well.

So it's GPT-O-S-S.com. Very simple. It is a chat G-P-T-esque interface. You don't have to have an account. You don't even have to log in. You choose which model you want. Do you want the 20B, the 120B? Do you want high, medium, or low reasoning?

That's it. And then you can go really try this thing out. go code with it, go have it write something, go have it be your brainstorming partner, have it do research on emerging markets, right? This is where you really have to have your company's internal benchmarks in play, because this is a huge opportunity for cost savings. And I still know there's a lot of companies out there that are using just one model. So as an example, I know there's a lot of companies.

out there that are using like a Claude 3-5 sonnet or a Claude 4 sonnet because someone on the development side needed it. But then they use it on the marketing side, which is terrible if you're using it on the API because it's so expensive, right? There's so many use cases that a model like this is just going to be better than a paid model, right? But you have to have all of your benchmarks ready, right? Maybe for software development, a model like GPT,

isn't going to do the trick. Maybe it will. Maybe it won't. But maybe for customer service, this will do the trick, right? So it's important you have your internal benchmarks ready to go because this is something that could work. Okay. Here's how this works, all right?

The actual open source nature. Because a lot of companies are wondering from a competitive perspective, like how does this add up? So let me explain. So opening I released the weight. they released the mixture of experts architecture and they released the inference code. But because a lot of people are like, okay, well, now what advantage does open AI have? If they just release this open source thing, isn't everyone else just going to copy and paste it and,

you know, essentially put out a better open model? Well, yes, so you can modify this, but it is still part of their model. But there are certain things on how they got to this point that's protected. Right. So the training data sets, you don't get that when you are using downloading, fine tuning on an open source model. Their data curation methods, their reinforcement learning with human feedback alignment techniques. Right.

So there's still a lot of things that are under the covers. So if you're running from a competitive standpoint, oh, if this is open source and people can download it, don't they just see every single thing? No, you don't. But competitors get the final product, essentially, but they can't reverse engineer OpenAI's competitive mode, which is their people. This is their engineers are the best in the world, right? Obviously, Google is giving them a run for their money now, but largely a smaller team from Open AI. But Google has been shipping faster.

So, well, actually, I'm excited to see how Google responds. Google's been dropping some cryptic tweets this week that they have a big week as well. We've already seen a couple releases from them. So it's been fun to watch them again, go back and forth. So what is Open AI doing? Right. They're making billions of dollars by charging people on the API end. And now they're essentially giving away.

It's not the most powerful model, right? It's not as powerful as their 03 or their O3 pro. But I would say in my use so far, I would peg this somewhere, and I know this is confusing alphabet soup. I would pig this somewhere depending on your use cases, but you can think of it as a GPT40 level to an 04 mini, right? Because it's reasoning. So I think for a lot of use cases, it falls kind of between there, which is a really good model. Right. But why?

Again, opening eyes going scorched earth. this is fun. It's fun for me as someone that covers AI every single day to see this happen because there's going to be so like entropic is going to be in the hot seat. Meta is going to be in the hot seat. All these mid-tier providers, they're going to be in the hot seat, right? But reports have said that OpenAI's enterprise market share on the API side has gone down as Chinese alternatives have gained steam.

Right. So about starting in. early 2025, a lot of the Chinese AI companies came out with some very powerful, very capable open source models. I wouldn't use them if you're a U.S. company, FYI, I've done plenty of shows on that, but I wouldn't touch them with a 10-foot pole. Anyways, I mean, companies like very capable models, not good to be sharing your data, especially on the website.

So Kimmy, Deep Seek, Quinn, very good models that had been pushing into that global market share. So I think what Open AI is doing here, it's a strategic choice. They are commoditizing the mid-tier market rather than starting to lose that to the competitors, right? A lot of these, I think there's so many companies, especially here in the U.S. that have been on the fence, right? Because they see these open source models, but they're from China. And there's a lot of data privacy and security issues with that that they want to, but they're not quite sure. And some have, right?

And some of the global share has gone, especially on the developer side, small developer, a solo developer, which, you know, they're making great, great pieces of software. So I think this is just opening I saying, yeah, they feel confident. Essentially, I think they only see Google as their only competitor. And they're like, we're going to put a powerful open source model that it's going to start to bankrupt some of our smaller competitors. Maybe this is a move. And they'll eventually acquire or acquire.

hire some of them because this is going to bleed some of them out because some of the AI labs cannot make a model even proprietary as good as this one. It's one of the benefits of being the company with the most users. You have the most user data, which also helps you build better models. So it is a strategic choice here. They're saying, all right, we'll be fine, losing a couple billion dollars this year because what we're going to do is we're going to bleed out the small guys we're going to get more people using our platform and i like to think of it it's like a sample right it's anyone like me like to go to costco on a on a on a saturday or sunday in man i can get a whole whole meal on a saturday or sunday at costco just crush it right um but i mean i end up buying

a bunch of stuff that i didn't go there to buy because i eat it and i'm like oh this is better than know, this, I don't know, this brownie is better than the brownie I've been buying or this, you know, energy bar is way better than what I normally eat, right? And then you switch over, right? You get a sample and you're like, oh, this is great. I love this. So that's, that's also, right, kind of a freemium offering or a freemium kind of onboarding ramp for, you know, enterprise clients as well, right? I think a lot of people are going to be jumping off Anthropic. A lot of people are going to be jumping off cohere. A lot of people are going to be jumping off co here. A lot of people are going to be jumping off mistral matter as well. I mean,

they're going to start and they're going to get in here and they're going to say, hey, for six of our eight use cases, this new open source, GBTOSS works. For the other two, well, since we're already on the open AI platform and we've kind of worked our processes around them, we're going to get GBT5 or we're going to get O3 on the API side. Right. So it's a smart move, but they're essentially saying, yeah, we're going to burn out. We're going to lose some money. But we're going to squash everyone. And in the long term, we're going to gain a lot. So in the short term, scorching the earth, medium long term, I think a huge game.

I think the only company that is going to be able to compete with this in the long term is Google. Winners and losers, I've already talked about them. But let's talk about a huge winner. My gosh, Nvidia. Like, just wait. And this is going to be a gradual realization, right? It's not like people are going to wake up in three days. and be like, oh my gosh, this is huge for NVIDIA.

It's just going to be a gradual thing over the next year or two. As we see that the type of edge AI, the type of on-device AI that's now possible because of this new model, right? Think of this model in the future being on an iPhone. Think of this model, you know, being on every single Windows laptop, right? And Microsoft did announce that they are going to be incorporating this into some of their hardware. It's amazing. This opens up so many capabilities, but you need the Nvidia chips. There's very few chips in the world that can handle this at scale.

Invidia, huge winner here, huge winner. I'll actually go, you know, I'm sure at some point in the coming weeks, I'll go look at Google search trends for certain consumer, you know, Nvidia GPUs. I'm sure they're spiking. The H-100 searches are going to be spiking, right? That's more for the enterprise side. here's a weird one.

I think Google's a winner here. Oh, yeah. Here's why. I think this gets the mainstream conversation going about putting capable large language reasoning models on a phone. Where probably a year or two off, Apple's too slow to do anything competent in the AI world. So I think this actually plays to Google's advantage,

because now all of a sudden, consumers, you know, businesses who need edge AI, maybe they weren't thinking about it. But this is going to drive, I think this is going to drive the conversation. This is going to be that pivot point. But guess who's already there? Google is. Google has a very capable Gemma 3 open model. So it's a little different.

It's a bigger drop off between, you know, Gemini 2. 2.5 Pro and Gemma 3. I'd say it's more of a mini Gemini where I'd say GPTOSS is a capable, right? It's a medium. It's a medium size, right? Like I said, it's probably in some instances between GPT40 and GPT 04, or sorry, 04 mini. So it's a very capable model where Gemma is, it's a step down, but it's still a great model. you have Gemma 3 already on smartphones, right?

And I think that Google's Gemma 4 could be at the same level as today's GBTOSS. Why does that matter? Yeah, I think Google's going to be caching in on the hardware side. And I think that this is actually going to push and popularize their Gemma series, which I think doesn't get enough love. I've gushed on it plenty from the time that it was first released. I'll have to go back and see, but I believe when it was released, right, like I know in boxing, people say, you know, it's pound for pound, the best fighter, right?

And it's usually, I don't know, some dude that weighs like 120 pounds that no one's ever heard of, right? But pound for pound, when Gemma 3 was released, it was the best model by parameter. And it wasn't even close, right? So it's a very lightweight model that punches way above its weight class. So I'm excited to see what happens with Gemma 4. And I think Gemma 4 could be the first mainstream edge AI kind of push. Losers, I kind of already talked about some of them.

I think mid-tier API providers, I think, you know, when we talk about kind of these tiers of the AI labs, the AI companies, right, you have Open AI and Google heads and shoulders above everyone else. they're like, you know, flip-flopping. We'll just say they're number one. We say, I like to say Microsoft and Anthropic. Again, you know, people, we can argue about this all day, but they're in the conversation. They're like the 1B tier. Anthropic might be in trouble, right?

So you're already hitting at the 1B tier. And then the 1C tier, they're in trouble. They're in trouble, right? So the mistrules, the coheres, right? There's so many of these labs that a lot of people maybe haven't heard of. Meta, right? This is going to hit Meta's user base as well. This is going to be a pretty big shakeup.

Because, again, you have to think, it's not just the model that they're putting out. There's going to be thousands. There probably already is thousands of fine tunes, even though the model has only been out for, I don't know, 36 or 72 hours or something like that. There's thousands of fine tunes. There's going to be so many industry-specific, domain-specific versions of that 20B GPT-OSS popping up overnight. And they're going to be really, really good. So let's talk a little bit more about the edge computing timeline. And I think that's big here.

And I don't want to overlook that. There's so many things that I want to talk about with GPT. OSS. But edge computing has to be one of them, right? I think that's the end game. The end game is we're not using. the cloud. The end game is the world's most powerful models are going to live on our laptops. They're going to live on our phones. They're going to live on our tablets. And that changes things,

right? Security becomes less of an issue because your sensitive data doesn't leave your device. It stays on device, right? That's, you know, on device AI, edge AI, kind of the same thing. but I do truly think that this, this in theory could save Apple. It could save Apple. If Apple gets their act together and if their iPhone 17 or iPhone 18 can handle this model, they could actually have AI, right? And they could avoid all these class action lawsuits for promoting AI and then not actually getting AI.

Did you, did you guys see the Google ad? Oh, so good. They didn't even mention Apple by name. It was something about like, hey, if you want a phone with AI, but not just a promise, like something that actually delivers. Great, great, great, great ad, by the way. But this is actually good news for Apple. And Apple could get their act together, right?

I'm sure the iPhone 17 is probably too late to change the specs. But the iPhone 18, Apple could actually have meaningful edge AI by 2030, which before this, I wouldn't have thought possible, right? even giving Apple five years, no. Again, this can change how we all live and work, right? Think of the best ever large language model experience you had, right? You're using a cloud provider, you're uploading all your data. It might take two, three, four, five, ten minutes.

Now think of that without the cloud. Think of that on your phone, right? Think of it like, hey, what was that picture I took and, you know, who was I texting about that? and what did the email say and it just knows instantly, right? And I know, you know, Google is a little ahead there and on some of the Samsung phones. A couple other things that we need to talk about, safety. Because what does this mean, right? We can think about all the good things.

But when you released a model like this without restrictions into the wild, it's great risk, great reward, right? we might be able to solve certain diseases, discover new medicines. There's a lot of downside. There's bad actors, right? So Open AI did say they conducted the first of its kind safety analysis intentionally maximizing bio and cyber capabilities just for this open source model. Also, malicious fine-tuning removes safety guardrails for disinformation, cyber attacks,

bio-weapons, research successfully. So a lot of work went in. opening. I did say a lot of work went in and they recognized that releasing a model this powerful can create a lot of bad and it can't. Bad actors will be able to bypass those things. These open source models will get jail broke, jail broken. Bad things will happen, right? But it's, I like to tell people, it's the same thing with the internet. The internet came out, you know, people, you know, bad, bad actors connected on the internet. They do bad things. on the internet. It's going to be the same thing with large language models on the cloud.

It's going to be the same thing with offline, but especially offline because you can't trace it. Once someone downloads it, that's it. You have no clue what they're doing with it anymore. You have no control. So there's zero ability to recall, update, or monitor usage once it's downloaded globally. But the big picture that I want to leave everyone with is this is the pivot point. I talked about this. This is bigger.

in GPG5. I think this is the strategic pivot going from closed AI, proprietary AI to open models. I think this is going to reshape the international AI race. In this right now, open source, capable open source models from the big boys, it forces all providers to accelerate innovation or lose relevance. If you thought AI innovation was fast in 2025, wait until the last four months here. It's going to go warp speed. Right now, it's U.S. and China competing. Who's going to get to AGI first? And I think one of those things, once you go open source, there's no going back.

It makes other open source models better. It makes proprietary models faster. No longer can companies just sit on a proprietary model for a year, which we've had reports of, oh, this model's been done for nine months, and they're just waiting to release it. They're waiting to see what their competitors do. That's not going to happen anymore. The pace is going to be. even faster as we race toward AGI.

We've talked about it first. AGI is the new gold plus power plus electricity plus currency times a million. If you get to AGI first, you can control everything. You become the global superpower. And actually open source is a big step to get there because it pushes innovation on both sides, all sides. Right. Because now all of a sudden, you're going to have researchers from Open AI and other labs, see what thousands of independent developers are going to do with this open source model to make it better, to make it more useful. Literally, I do think we're going to see it.

We're going to see these two models, these two open source models from Open AI. Like I said, bad things are going to happen, but they're going to cure diseases. They're going to discover new proteins. They're going to help with drug discovery. Literally, open source models. right so this is going to accelerate everything and be bigger than gpt5 it is we're going to hear today or you'll hear tomorrow if you're listening on the podcast i'll do a show on gpt5 um gpt5 will be a major model improvement but this is bigger gpt oss is a step change in the future of technology

and how business works because like i said AI labs can't sit on models anymore. They'll go out of business. Innovation is going to overdrive. And that changes what's possible. For me, you and business leaders everywhere. So now more than ever, you got to pay attention. You got to keep up to date and you have to put it all into practice.

All right. So if you miss anything, we're going to be recapping it in the newsletter. If you haven't already, please go to your everyday AI.com. Sign up for the free daily newsletter. Join us tomorrow. We're going to be going over the GPT5 release. So thank you for tuning in. Hope to see back tomorrow for that and every day for more everyday AI.

Thanks y'all. Meet Firefly AI Assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerate. rates execution. Stand control with the ability to step in and refine at any time. See it today at

firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.