WEBVTT
00:00:01.040 --> 00:00:04.320
What if you could make a billion people smarter?
00:00:04.559 --> 00:00:07.280
Yeah, it's kind of what OpenAI just did.
00:00:07.440 --> 00:00:30.800
Uh, there's usually never been a free AI plan where I actually think it was good enough or reliable enough for everyday business use, but OpenAI just announced that as they quietly kind of announced that free users, so users on the free plan, get unlimited use of GPT 5.6 Luna, a really good and capable model.
00:00:31.039 --> 00:00:40.000
So previously, free Chat GPT users were usually served a combination of GPT 5.5 and GPT 5.4, depending on the thinking level they chose.
00:00:40.560 --> 00:00:45.439
Actually, free and low price AI really had a week in the limelight.
00:00:45.600 --> 00:00:57.280
So between that announcement from OpenAI, there was also a new open weight powerhouse model from Alibaba and a new model from Meta that's basically free if you let them train on your data.
00:00:57.600 --> 00:00:59.920
So you probably missed some of it.
00:01:00.000 --> 00:01:08.640
That's because between all the big model releases, there's actually dozens of smaller feature updates each week that can move the needle for your business.
00:01:08.799 --> 00:01:10.799
And if you miss one or two, it's no big deal.
00:01:10.879 --> 00:01:16.480
But if you really don't pay attention to the movements that happen between the big splashes, you'll miss out completely.
00:01:16.640 --> 00:01:24.640
And that's why on Fridays, we bring you the Friday features, our weekly update of the latest AI updates that you can actually use today.
00:01:24.879 --> 00:01:27.680
These aren't news or leaks or coming soons.
00:01:27.920 --> 00:01:31.840
These are just new AI features you can use this Friday.
00:01:32.000 --> 00:01:33.840
So let's get into it.
00:01:34.319 --> 00:01:36.959
So on today's show, here's what you're gonna learn.
00:01:37.040 --> 00:01:41.840
You're gonna learn how you can get essentially free frontier coding, or at least for free 99.
00:01:42.159 --> 00:01:55.760
If you let a certain company named Meta train on your data, you're gonna know what new open source standard there is for better agent orchestration and how basically all the big players in the game got in on it, except for one.
00:01:56.000 --> 00:02:02.640
And you're gonna know the new updates that give free Chat GPT users the best deal in AI right now.
00:02:02.799 --> 00:02:03.840
Let's jump into it.
00:02:04.000 --> 00:02:05.280
Welcome to Everyday AI.
00:02:05.439 --> 00:02:07.599
My name's Jordan Molson, and we do this every day.
00:02:07.760 --> 00:02:16.879
This is your daily live stream, podcast, and free daily newsletter, helping business leaders like you and me keep up with these AI updates because they quite literally don't stop.
00:02:16.960 --> 00:02:21.120
I tell you what's good, what's not, how to use it to grow your company and your career.
00:02:21.280 --> 00:02:30.000
So if that's exactly what you're trying to do, do me a favor, stranger, click that subscribe button on Apple Podcasts or Spotify, wherever you're listening.
00:02:30.080 --> 00:02:33.680
Uh, and then make sure you go to our website at your everydayai.com.
00:02:33.759 --> 00:02:34.479
It's in the show notes.
00:02:34.560 --> 00:02:37.759
So make sure you go check that out and sign up for our free daily newsletter.
00:02:37.840 --> 00:02:39.919
We're gonna be recapping the highlights from today's show.
00:02:40.000 --> 00:02:45.599
So if you miss anything, don't worry, it's in there as well as all the other AI updates you need to know to get ahead.
00:02:45.759 --> 00:02:49.280
All right, let's get into it and look live.
00:02:49.439 --> 00:02:59.120
All right, so yeah, if you are watching uh or sorry, if you're listening to this, there's always a video version, not too much that you need to see uh today, if I'm being honest.
00:02:59.360 --> 00:03:04.879
But um you can always check that out on our website at your everydayai.com.
00:03:05.039 --> 00:03:06.319
All right, here we go.
00:03:07.520 --> 00:03:09.199
Eventually, come on, screen.
00:03:09.280 --> 00:03:09.840
There we go.
00:03:09.919 --> 00:03:11.520
Yeah, it's unedited, folks.
00:03:11.680 --> 00:03:14.080
All right, our first big AI update.
00:03:14.240 --> 00:03:16.159
If you're creative, you're gonna like this one.
00:03:16.240 --> 00:03:22.639
That's because Adobe just launched a big update to its unified plugin in Chat GPT.
00:03:22.800 --> 00:03:32.400
Now it can access more than 70 tools from Photoshop, Firefly, Acrobat, Express, Premiere, Lightroom, Illustrator, InDesign, and Stock.
00:03:32.719 --> 00:03:34.400
So essentially, how does this work?
00:03:34.479 --> 00:03:39.039
Well, you describe what you want it to do, and it picks the right Adobe tools automatically.
00:03:39.199 --> 00:03:45.840
So you can edit photo sets, make campaign videos, uh, customize different design templates, turn spreadsheets into PDFs.
00:03:45.919 --> 00:03:51.520
So it kind of replaces the separate Adobe apps that were available in ChatGPT before.
00:03:51.599 --> 00:03:59.360
So now it's just one unified app where you don't even have to like know, oh, should I be using you know Acrobat for this?
00:03:59.520 --> 00:04:03.599
Should I be using uh, you know, Lightroom, Photoshop?
00:04:03.680 --> 00:04:04.400
It doesn't matter.
00:04:04.479 --> 00:04:06.080
So you just open Chat GPT.
00:04:06.159 --> 00:04:14.560
Uh, if you do have that Adobe Creative Crowd, uh Creative Cloud subscription, and you just dump in a bunch of files and say, do all this stuff to them, and it does it.
00:04:14.639 --> 00:04:19.680
Uh, so Adobe is running this play on every chat bot, not just Chat GPT.
00:04:19.759 --> 00:04:26.079
Uh, its tools are already on Claude and Copilot, but I believe this is the first um in terms of these types of features.
00:04:26.240 --> 00:04:28.879
This is the first one uh we've seen that's this robust.
00:04:29.040 --> 00:04:30.079
So who has access?
00:04:30.240 --> 00:04:34.160
Well, anyone, if you are on Chat GPT, you got it.
00:04:34.240 --> 00:04:39.839
And there's also four um mini uh features that are in included in this announcement.
00:04:40.000 --> 00:04:42.800
You don't need a plan uh to use them.
00:04:43.120 --> 00:04:47.439
So yeah, even right now, you don't need an Adobe account to start.
00:04:47.600 --> 00:05:02.240
Even just using guest mode does give you a broad set of tools immediately, not everything, because signing in with your Adobe account does unlock uh some of the more generative, uh, you know, heavy AI features, uh, and then being able to sync everything in your creative cloud.
00:05:02.319 --> 00:05:08.879
So it also does work inside of ChatGPT work in codecs so it's not just on the uh ChatGPT.com release.
00:05:09.120 --> 00:05:10.879
So who's this useful?
00:05:11.040 --> 00:05:11.920
Well, anyone.
00:05:12.000 --> 00:05:18.160
If you're trying to do something creative, if you are a creative, if you're not creative, but you want to be creative, this is a great way to do it.
00:05:18.240 --> 00:05:32.800
So obviously, open AI and ChatGPT has a fantastic, literally state of the art um kind of their uh GPT images too, which is really, really good, but it can do everything that you might want it to.
00:05:32.959 --> 00:05:33.600
Um, right.
00:05:33.680 --> 00:05:38.240
So in those cases, now you have this tool from uh Adobe.
00:05:38.399 --> 00:05:40.319
All right, so that's a good one.
00:05:40.480 --> 00:05:48.639
But here's uh maybe even a better one, maybe for a lot of our audience, because meta, another pretty big splash here.
00:05:48.800 --> 00:05:58.800
And there's actually a plan here that I wasn't expecting from meta, and I don't hate it as one that I'm just like, hey company, take my data.
00:05:59.040 --> 00:06:06.079
I don't hate this new pricing from Meta Muse, Spark, and their new Muse Code and Spark 1.2.
00:06:06.160 --> 00:06:11.759
So here's what's new Meta released Muse Code, which is their kind of take on Claude Code or Codec.
00:06:11.920 --> 00:06:17.920
So that's a terminal coding agent right now, uh, powered by its new Muse Spark 1.2 model.
00:06:18.000 --> 00:06:19.360
So, yeah, two different things.
00:06:19.439 --> 00:06:23.199
Uh, it is Muse Code, so that's a terminal coding agent, right?
00:06:23.279 --> 00:06:35.439
So, not like the desktop version of Cloud Code or Codecs, like the CLI version of Cloud Code or Codex, and then uh the next one is the new model, uh Muse Spark 1.2.
00:06:36.000 --> 00:06:42.319
So the pricing here, even if you're paying the full API pricing, pretty good rates in terms of performance.
00:06:42.480 --> 00:06:48.720
It's a dollar 25 per million tokens in and four dollars and 25 cents per million tokens out.
00:06:48.959 --> 00:06:54.879
But there is a new what they're calling contributor tier, it's an interesting name for it.
00:06:54.959 --> 00:07:04.240
Um, that drops that price down to 10 cents in and 20 cents out if you let Meta train on your data.
00:07:04.399 --> 00:07:10.959
So that's roughly 95% off in exchange for essentially your code, right?
00:07:11.199 --> 00:07:21.759
So um at first I'm like, absolutely no one's gonna do this, but then I'm like, yeah, like I don't know, especially if you're starting from scratch, uh, right?
00:07:21.839 --> 00:07:30.720
Let's just say that you have a personal project, or you know, if you're an indie hacker, indie developer, I'm like, that's actually not a terrible idea, right?
00:07:30.879 --> 00:07:36.879
Obviously, if you're working with large proprietary anything enterprise, no one would ever touch that contributor tier.
00:07:36.959 --> 00:07:43.680
But I'm like, for everyone else, that it's more of a hobbyist, it's actually, I think, not that bad of a deal.
00:07:44.160 --> 00:07:50.720
So um, every model call, tool run, approval, and edit uh does get written to a local event log.
00:07:50.800 --> 00:07:54.319
So if the agent crashes, it can resume where it stopped.
00:07:54.480 --> 00:07:55.920
So who has access to this?
00:07:56.000 --> 00:07:58.000
Well, it's open now in beta.
00:07:58.240 --> 00:08:01.759
You can literally just go run a terminal command and you're off and running.
00:08:01.920 --> 00:08:17.519
Uh so the new model though, meta spark uh sorry, Muse Spark 1.2, is available inside Muse Code and to any developer through the meta uh meta model API, and it does have a 1 million token context window.
00:08:17.920 --> 00:08:20.079
So uh why is this useful?
00:08:20.240 --> 00:08:30.480
Well, aside from writing code, if you need persistent background sub agents that can work on hard tasks on their own and decide when to report back, uh it's pretty good for that.
00:08:30.639 --> 00:08:37.759
Uh, you know, the scores so far um for what's available on artificial analysis as an example, pretty good, right?
00:08:38.000 --> 00:08:46.639
You know, all of a sudden, meta went from not very relevant for like a full year and a half to now all of a sudden they're in the conversation, right?
00:08:46.799 --> 00:08:51.759
They're again Google might change this, you know, any any minute now, right?
00:08:51.840 --> 00:08:52.559
Watch Google.
00:08:52.639 --> 00:09:02.480
There's been uh rumors that we're getting something, you know, maybe this week or August 10th, uh, either Gemini 3.5 Pro or if they skip straight to Gemini 4 Pro, who knows?
00:09:02.639 --> 00:09:06.159
But Meta is actually ahead of Google, right?
00:09:06.240 --> 00:09:10.080
Not just in terms of model capabilities, but on pricing as well.
00:09:10.240 --> 00:09:15.200
So Meta has had a very impressive uh last month or two.
00:09:15.360 --> 00:09:33.919
So I think if you're you know a dev working on a large uh repo, if you want long running agents, like I said, or if you're an indie dev or a cost-sensitive team and you're not necessarily working on you know proprietary data, uh that contributor tier would just still a funny name for it.
00:09:34.080 --> 00:09:35.039
Not bad, right?
00:09:35.200 --> 00:09:40.799
Uh, we'll see if Muse code uh goes for very long, right?
00:09:41.039 --> 00:09:44.559
Um, I don't know, maybe this is the non-developer in me speaking.
00:09:44.639 --> 00:10:03.039
I've always been able to do the basics in a terminal, but I feel even the people that have been hardcore, like, hey, I'm a terminal coder, you know, in in terms of being a software engineer or dev, I've seen even some of the people that, you know, a year ago were like, I'm never giving up the terminal to being like the terminal just doesn't make sense anymore.
00:10:03.200 --> 00:10:03.919
So I don't know.
00:10:04.080 --> 00:10:04.559
We'll see.
00:10:04.639 --> 00:10:16.480
I mean, you have to start somewhere with uh, you know, in terms of having a coding agent, but regardless, meta's showing where they're gonna play, they're gonna try to compete with anthropic in Google on token pricing.
00:10:16.559 --> 00:10:18.960
I don't think that's necessarily OpenAI's game.
00:10:19.039 --> 00:10:22.879
Uh, that is where uh Anthropic reportedly gets about 80% of its revenue.
00:10:22.960 --> 00:10:32.000
So this to me looks like a shot at Anthropic, um, you know, bringing the actual coding agent with views code, uh, extremely cheap and affordable.
00:10:32.240 --> 00:10:34.240
And the performance is there.
00:10:34.879 --> 00:10:35.360
All right.
00:10:35.679 --> 00:10:37.200
Our next piece of AI news.
00:10:37.360 --> 00:10:40.720
Speaking of, well, very affordable uh models.
00:10:40.879 --> 00:10:45.919
Well, we have a new uh not quite king of the open hill.
00:10:46.000 --> 00:10:55.519
It's kind of uh actually let me check that because I know artificial analysis just put out a new version um of their four point, what is it, four point one one now?
00:10:55.600 --> 00:10:58.720
So I'm I'm double checking why to see uh okay, yeah.
00:10:58.879 --> 00:11:04.879
So they are just a little bit behind Kimi K3, uh, but essentially uh it is in the same tier.
00:11:04.960 --> 00:11:10.000
So Alibaba's new uh open weights Quinn 3.8 Max.
00:11:10.080 --> 00:11:11.840
Um, it is in the same tier.
00:11:11.919 --> 00:11:20.879
So it is in that uh, you know, GPT 5.6, uh Claude Fable 5, Claude Opus 5, Kimmy K3, and Quen 3.8.
00:11:21.120 --> 00:11:23.840
That's kind of your top five models in the world now.
00:11:23.919 --> 00:11:29.120
You know, we have two open weights models from China uh in spots four and five.
00:11:29.279 --> 00:11:38.799
So here is what it is it is Quen 3.8 Max, a 2.4 trillion parameter open weight model with a million token context window.
00:11:38.879 --> 00:11:41.120
It takes text, images, and video.
00:11:42.080 --> 00:11:46.320
Um, so here's the um interesting thing.
00:11:46.480 --> 00:11:51.440
So it did become uh the top-ranked Chinese model for text only.
00:11:51.600 --> 00:11:53.759
All right, so for text only, it takes that.
00:11:54.000 --> 00:12:11.279
And here's the thing, though, there was a report literally hours ago uh from Reuters that said Alibaba plans to begin charging major enterprise users for Quen 3.8 Max to charge them a share of their revenue they generate starting next week.
00:12:11.440 --> 00:12:13.039
So that one's interesting.
00:12:13.200 --> 00:12:15.120
So who has access right now?
00:12:15.200 --> 00:12:16.320
Uh well, anyone.
00:12:16.399 --> 00:12:27.200
So if you're working on the hosted API or through Alibaba Cloud Model Studio, uh so yeah, if you are in Alibaba Cloud Accounts, you can use it right now.
00:12:27.600 --> 00:12:30.639
Uh and the API is OpenAI compatible.
00:12:30.720 --> 00:12:34.960
So switching is a base URL and model uh ID change.
00:12:35.039 --> 00:12:40.480
So it's open weights, although the weights are not out yet, uh, but they did promise them within days of the release.
00:12:40.559 --> 00:12:42.559
That's essentially what Kimmy did as well.
00:12:42.639 --> 00:12:46.000
Um, and they kind of hit their target deadline for releasing the weights.
00:12:46.159 --> 00:12:50.960
But when the weights land, obviously you can't like self-host this.
00:12:51.039 --> 00:12:51.759
Um, right.
00:12:51.919 --> 00:13:01.120
And again, unless you happen to have a data center, uh, you know, a bunch of GPUs that you just weren't using in your basement, you aren't gonna use this, right?
00:13:01.200 --> 00:13:08.480
It's kind of this um uh you know, I don't know, fortunate, unfortunate uh current dichotomy of open weight open source software, right?
00:13:08.639 --> 00:13:12.080
I think a year ago, the best open models you could, you know, download.
00:13:12.159 --> 00:13:14.879
If you had a big expensive Mac, you could do it.
00:13:14.960 --> 00:13:16.000
You can't anymore.
00:13:16.080 --> 00:13:26.960
Uh, right, these multiple trillion parameter uh you know open weight models, they're not really for running locally anymore unless you literally have a data center.
00:13:27.440 --> 00:13:29.120
So, well, why is this useful?
00:13:29.200 --> 00:13:32.720
Well, it's frontier level capability with open weights.
00:13:32.879 --> 00:13:39.600
Um, so yeah, if for bigger companies, you can, you know, fine-tune these models, or if you're working with a provider, you can fine-tune them.
00:13:39.679 --> 00:13:44.639
So, unlike proprietary models, uh, where it's not always as easy to do something like that.
00:13:44.879 --> 00:13:53.360
Um, so uh also interesting to note, it did debut number four on the front-end code arena.
00:13:53.600 --> 00:13:59.440
Uh, so you know, apparently it has some pretty good front-end uh taste to it as well.
00:13:59.519 --> 00:14:01.200
So, who's gonna find this valuable?
00:14:01.360 --> 00:14:04.080
I think really just large enterprises.
00:14:04.240 --> 00:14:04.639
That's it.
00:14:04.799 --> 00:14:16.960
So if you're hedging against, you know, US lab pricing and lock-in, uh, or just you know, people who are already self-hosting large models, researchers, uh, they are coming out with a 27B version.
00:14:17.200 --> 00:14:25.120
Uh, you know, so if you do have uh, you know, a couple heavy Mac studios beefed together, you can run it like super slow or something like that.
00:14:25.279 --> 00:14:31.120
Uh actually 27B, yeah, you should be able to run it on like the most expensive DGXs.
00:14:31.279 --> 00:14:35.759
Um, but yeah, again, nice open model, uh, open weight model.
00:14:35.919 --> 00:14:43.200
So if nothing else, uh it's pushing the other open models, which are pushing proprietary models, which are push pushing prices down, right?
00:14:43.279 --> 00:14:45.519
That's ultimately what's happening here.
00:14:45.600 --> 00:14:58.639
Uh, because the people who can afford uh to run something like a Quen uh Quen 3.8 Max by running it locally are those, you know, Fortune 500 types that prep that may have access to compute.
00:14:58.799 --> 00:15:06.639
So that is the people who are generally, you know, using anthropics models would probably be their first customer, OpenAI, Google maybe as well.
00:15:06.799 --> 00:15:14.000
So, you know, that's also going to push uh in theory, you know, either push down the prices from the major frontier labs or at least keep them in check.
00:15:14.080 --> 00:15:18.879
So even if you're like, okay, this model isn't gonna impact me or my business, we can't run it.
00:15:19.039 --> 00:15:20.159
Well, you can, right?
00:15:20.240 --> 00:15:22.639
There's always you know providers that you can use.
00:15:22.799 --> 00:15:26.799
Um, but if anything else, keep models in check.
00:15:26.960 --> 00:15:28.559
All right, let's go next.
00:15:28.720 --> 00:15:34.399
Uh, next, we already covered this one in depth on our Wednesday AI Working Wednesday show.
00:15:34.480 --> 00:15:36.320
That's episode 834.
00:15:36.720 --> 00:15:40.000
But a ton new with Gemini Notebook.
00:15:40.080 --> 00:15:48.639
Yes, the tool formerly known as Notebook LM finally got its big glow up that rolled out to all paid users.
00:15:48.799 --> 00:15:54.080
So this is technically an older update, but we didn't cover it on the show previously.
00:15:54.159 --> 00:15:58.480
Well, aside from Wednesday, uh, when it was, you know, I think it was just released on Tuesday.
00:15:58.639 --> 00:16:02.320
We didn't really cover it before because it was only available to ultra subscribers.
00:16:02.480 --> 00:16:09.600
So just so you know, on these Friday shows, these are just you know tools or updates that are available to your standard, you know,$20 base plans.
00:16:09.679 --> 00:16:14.320
But now we have that Gemini Agenc notebook that is out.
00:16:14.480 --> 00:16:16.000
So yeah, has a new name.
00:16:16.240 --> 00:16:22.879
Notebook LM is gone, Gemini Notebook is here, and Gemini Notebook, at least for paid users, is agentic by default.
00:16:23.039 --> 00:16:29.120
So Google finished rolling out the agentic version of Gemini Notebook, the upgrade to pro subscribers.
00:16:29.279 --> 00:16:39.919
So every notebook now gets a secure cloud computer that writes and runs real code against your sources, plus auto-generated charts, spreadsheets, slide decks, and PDFs.
00:16:40.080 --> 00:16:47.840
So it runs on Gemini 3.5, but it also uses the anti-gravity harness and it has a hundred built-in skills.
00:16:48.080 --> 00:16:52.080
So think of like code interpreter grounded in your documents.
00:16:52.240 --> 00:16:59.919
Uh, you can also just, you know, start a blank notebook, give it a vague query, and it can also now use the web, although it will confirm with you, right?
00:17:00.000 --> 00:17:05.519
Because that's the good thing about, you know, previously Notebook LM is it stayed completely grounded in your source data.
00:17:05.680 --> 00:17:12.240
So now uh Gemini Notebook can query the web, uh, which is a little bit different for me, right?
00:17:12.400 --> 00:17:19.359
But it will at least let you know hey, you know, if you're asking about something that's not in your source material, it will tell you I don't have the answer.
00:17:19.519 --> 00:17:21.440
Do you want me to go search the web?
00:17:22.240 --> 00:17:24.720
Uh so uh who has access?
00:17:24.960 --> 00:17:28.559
Anyone with a uh Gemini Pro account?
00:17:28.799 --> 00:17:30.400
Uh, you've got it now.
00:17:30.480 --> 00:17:32.799
Uh so this is on the web version only.
00:17:32.960 --> 00:17:35.680
If you're an ultra subscriber, you already have this.
00:17:35.839 --> 00:17:40.240
Uh free users, you're still gonna see Gemini Notebook, right?
00:17:40.319 --> 00:17:45.599
Uh, but you're not gonna have these new agenc the code execution or some of these new output formats.
00:17:45.680 --> 00:17:46.960
So that's the other thing that's new.
00:17:47.039 --> 00:17:53.599
Uh in Gemini Notebook, you can output as a PDF, uh, a doc, um, you know, different PNGs, uh, right.
00:17:53.680 --> 00:17:58.160
So essentially before you were a little more limited in what you could actually output.
00:17:58.240 --> 00:18:02.880
Uh, there was, you know, I think about eight different sources on the right hand side in the studio panel.
00:18:03.039 --> 00:18:05.839
Now you can just say, you know, output in these different sources.
00:18:05.920 --> 00:18:09.519
So I actually did what I thought was a pretty good rundown on Wednesday.
00:18:09.599 --> 00:18:12.079
So if you do want to check that out, make sure you go do that.
00:18:12.240 --> 00:18:13.680
Episode 834.
00:18:13.839 --> 00:18:15.359
Who's gonna find this valuable?
00:18:15.599 --> 00:18:17.440
Literally anyone, right?
00:18:17.599 --> 00:18:24.720
Um, if you're you know a heavy uh chat GPT work, codex, claud code, uh user, right?
00:18:24.799 --> 00:18:28.559
As an example, these new agentic features are pretty good, right?
00:18:28.720 --> 00:18:42.960
Um, I'm really looking forward to when um you know Google does introduce whether it's Gemini 3.5 or Gemini 4 Pro, whichever one, and then seeing those roll out to Notebook LM, I'll be really excited for because they're technically running the Gemini 3.
00:18:46.160 --> 00:18:50.000
But the Gemini 3.5 flash model is not that great.
00:18:50.160 --> 00:18:54.079
Um, so yeah, it's got like a 30.
00:18:54.799 --> 00:18:55.359
What is it?
00:18:55.519 --> 00:18:58.319
Mid no, that's Gemini 3.6 flash.
00:18:58.400 --> 00:18:59.039
Come on, come on.
00:18:59.119 --> 00:19:01.279
Where's where's my Gemini 3 flash?
00:19:01.440 --> 00:19:03.440
Uh let's see, let's see where we have it.
00:19:03.599 --> 00:19:06.640
It is okay, a 52 on the artificial analysis.
00:19:06.799 --> 00:19:08.960
Uh, so not that good.
00:19:09.119 --> 00:19:15.680
So it is below, um, or no, sorry, it is at the same level as uh GPT-5.6 Luna.
00:19:15.759 --> 00:19:18.960
So okay, from that respect, it's okay-ish, right?
00:19:19.039 --> 00:19:25.519
So if you think that open AI has three tiers, um, you know, they have uh Seoul, Terra, and Luna.
00:19:25.680 --> 00:19:32.480
Right now, Google's best tier is the lowest tier for um OpenAI's Chat GPT.
00:19:32.559 --> 00:19:36.000
So take that for uh you know what it's worth.
00:19:36.160 --> 00:19:41.680
So not the best model in the world um that's powering it, especially when you're a paid user, right?
00:19:41.759 --> 00:19:48.640
We're gonna talk about the free uh kind of upgrade for uh uh GBT56 Luna, which I think is great for free users, right?
00:19:48.720 --> 00:20:01.440
But for being a paid user right now, it's kind of not a great time uh to be a you know Gemini first company, but I will expect Google any day, any minute, any week now to come out with a good 3.5 pro or Gemini 4 Pro.
00:20:01.680 --> 00:20:04.160
And I hope it gets rolled out to Gemini Notebook.
00:20:04.319 --> 00:20:10.240
All right, next we have a new open standard for agent plugins, and this is pretty cool.
00:20:10.400 --> 00:20:13.279
So here's what's new, and yes, it's available now.
00:20:13.359 --> 00:20:19.839
There's a new open vendor neutral standard for packaging AI agent extensions.
00:20:20.000 --> 00:20:26.960
So now it's a one plugin, uh folder with a plug-in JSON file bundling and agent skill and mcp servers.
00:20:27.119 --> 00:20:35.839
And right now it works across ChatGPT, uh, codecs, cursor, Google, GitHub, Copilot, uh, VS Code, and others.
00:20:35.920 --> 00:20:41.200
So essentially, you can build an agent plugin once and run it anywhere.
00:20:41.839 --> 00:20:45.680
So this is a little confusing, but let me just break it down.
00:20:45.759 --> 00:20:53.279
Uh, and hopefully this is like 99% correct because I haven't actually gone out and built this because it's only an hour like in a couple hours old.
00:20:53.440 --> 00:20:56.559
Um, so previously think you have skills, right?
00:20:56.640 --> 00:21:00.720
And skills are essentially just a bunch of text files in a folder, uh, right.
00:21:00.880 --> 00:21:09.599
And then you also have things like plugins, and plugins are you know multiple skills or a skill in an app together, um, right?
00:21:09.759 --> 00:21:12.240
And then you have like JSON files.
00:21:12.400 --> 00:21:15.920
Um, so it's kind of like the new agent plugins.
00:21:16.000 --> 00:21:20.079
Um, you know, and then you know, you have your skills are in markdown files, right?
00:21:20.160 --> 00:21:24.640
It's essentially pulling all of these things together, but in an open standard, right?
00:21:24.720 --> 00:21:27.519
So that's great because you had the skills MD.
00:21:27.680 --> 00:21:34.799
Uh you know, skills are pretty universal, um, right, because they are just text files, they're markdown files.
00:21:34.880 --> 00:21:36.799
Um, so those are pretty universal.
00:21:36.960 --> 00:21:41.759
Uh, also inside the uh agent plugins, sorry, I forgot to mention this, it's MCPs as well.
00:21:41.920 --> 00:21:51.279
So it's essentially stacking, you know, your skills, your MCP servers, um, you, you know, your JSON files, uh, you know, all in one package.
00:21:51.440 --> 00:21:58.559
So this is nice uh because before you would have to load all of these things in individually, uh, which takes a lot of time.
00:21:58.720 --> 00:22:03.599
And when you you know it's Slightly different across different providers.
00:22:04.160 --> 00:22:09.039
You don't always have that from a business strategy perspective.
00:22:09.200 --> 00:22:12.079
You get a little locked in at times, which you don't want.
00:22:12.240 --> 00:22:18.960
So this is why it was great to see almost every single big name instantly support the Asian plugins, right?
00:22:19.119 --> 00:22:21.359
Anthropic, obviously not on the list.
00:22:21.839 --> 00:22:25.839
I don't know why Anthropic just won't get on board with this.
00:22:26.079 --> 00:22:30.000
Um, you know, obviously they have their Claude.md, right?
00:22:30.160 --> 00:22:36.720
But it's like, okay, Anthropic, just change it to Agent MD, like literally every other company, right?
00:22:36.880 --> 00:22:40.799
Google, Microsoft, Cursor, OpenAI, everyone else is doing this.
00:22:40.960 --> 00:22:42.160
Anthropic, get on board.
00:22:42.319 --> 00:22:46.720
It's just making uh it's I think it's slowing the pace of AI down.
00:22:46.799 --> 00:22:49.680
Uh, I think it's a maybe a bad look on anthropic.
00:22:49.759 --> 00:22:52.480
Like, are you just trying to lock people in, right?
00:22:52.559 --> 00:22:55.200
Maybe I don't know, maybe they are, maybe that's part of their business strategy.
00:22:55.359 --> 00:23:03.599
But for me, right, I love uh kind of having this open, uh, you know, new agent plug-in standard because it does make it easier for businesses.
00:23:03.759 --> 00:23:13.119
Uh, the like the other thing, too, is um, yes, in most cases, I think it's best for companies to kind of pick one AI kind of operating system and stick there.
00:23:13.279 --> 00:23:17.680
But you know, you might maybe your finance team needs a different one, right?
00:23:17.839 --> 00:23:27.759
Maybe 90% of your company can be on uh you know Chat GPT work, but maybe 10% of your company needs to be on Gemini for whatever reason, right?
00:23:27.839 --> 00:23:34.799
And in this case, when you have these ancient plugins, it's very easy uh to have that standard and to be able to transfer that over.
00:23:34.960 --> 00:23:40.000
But you know, if anthropic doesn't get on board, it's not gonna be as easy for them, anyways.
00:23:40.319 --> 00:23:43.440
So openAI announced this, but it's not open AI's.
00:23:43.519 --> 00:23:51.920
So Vercell initiated it, uh, as well as uh Amazon, any sphere, which is cursor, GitHub Copilot, Microsoft, uh, etc.
00:23:52.240 --> 00:23:54.960
So um who has access right now?
00:23:55.039 --> 00:23:55.759
Well, anyone.
00:23:55.839 --> 00:23:57.519
So you can just build this, right?
00:23:57.599 --> 00:24:00.559
You can just point your coding agent to this uh the website.
00:24:00.640 --> 00:24:04.799
The website is just agent-plugins.org, right?
00:24:04.960 --> 00:24:06.400
All of the documentation is there.
00:24:06.480 --> 00:24:12.240
So this is just an easy way, especially if your team is becoming more agentic in your work, right?
00:24:12.319 --> 00:24:17.599
You have multiple MCP files, you have multiple skills, you have JSON files holding it all together.
00:24:17.839 --> 00:24:21.680
It's just an easier way to port this across server to server.
00:24:21.839 --> 00:24:23.759
All right, or I'm sorry, service to service.
00:24:23.920 --> 00:24:24.559
All right, next one.
00:24:24.720 --> 00:24:26.160
Speaking of being open, right?
00:24:26.240 --> 00:24:28.880
It's a big week for open and free.
00:24:29.039 --> 00:24:31.599
Uh, this one was not expecting this.
00:24:31.759 --> 00:24:36.640
Cloudflare announced their open source Cloudflare OS.
00:24:36.960 --> 00:24:41.839
Uh, it's an internal AI workspace that it built to actually run its own company.
00:24:41.920 --> 00:24:45.920
So they essentially said, we build this to run our company, so everyone else can have it as well.
00:24:46.079 --> 00:25:00.319
So uh essentially, every employee gets a browser-based HHID workspace with governed access to internal systems, research documents tied to live data, automated workflows, and shareable internal apps with well, no developers really need it.
00:25:00.400 --> 00:25:07.119
Uh, so thousands of Cloudflare's own employees across every team already use it daily, according to the company.
00:25:07.279 --> 00:25:11.039
So seems pretty battle-tested, not just like some demo software.
00:25:11.200 --> 00:25:24.799
So it competes technically, right, with Chad GPT Work or Cloud Co-Work or Gemini Enterprise, but it's free, open source, and runs in your own Cloudflare account instead of the kind of per seat SaaS.
00:25:25.279 --> 00:25:28.000
So uh who has access right now?
00:25:28.240 --> 00:25:30.799
Well, it's open source, so anyone.
00:25:30.880 --> 00:25:32.880
So you can go download the repo.
00:25:32.960 --> 00:25:41.920
Uh, you do obviously need a Cloudflare account to run it because it deploys into your own account uh using your own access policies and integrations.
00:25:42.000 --> 00:25:46.880
Uh, but your IT team, uh, essentially, yeah, they'll probably need to help you a little bit.
00:25:47.039 --> 00:25:56.240
Uh, but there is also a managed version through Cloudflare plus implementation partners uh that um are coming, but no date is given yet.
00:25:56.480 --> 00:25:59.839
So uh why is this useful?
00:26:00.000 --> 00:26:07.680
Well, you can just essentially bring your own model um and have an open source agentic platform to work on.
00:26:07.839 --> 00:26:14.720
So, yes, uh, this isn't free uh to use, it's just a free operating system.
00:26:14.799 --> 00:26:24.319
Think of it that way, but you still have to bring your apps, aka, you still have to bring your own model through the Cloudflare AI model gateway.
00:26:24.480 --> 00:26:28.640
Um, so but there's you know, you can do spend caps and you're not locked in.
00:26:28.720 --> 00:26:29.920
So think of it more like that.
00:26:30.000 --> 00:26:33.200
Think of it, you know, and there are other versions of this, right?
00:26:33.279 --> 00:26:40.400
Uh, there are versions of open, you know, open versions of um, you know, desktop agents, right?
00:26:40.480 --> 00:26:45.839
But this is from Cloudflare, one of the literally biggest names uh in the internet ever.
00:26:46.000 --> 00:26:48.160
Uh, so a very trusted source.
00:26:48.319 --> 00:26:54.079
So uh, you know, a lot of companies run their uh you know web operations through Cloudflare.
00:26:54.319 --> 00:26:55.920
So trusted, trusted name.
00:26:56.000 --> 00:26:59.519
Uh love this because it is giving people more options.
00:26:59.680 --> 00:27:06.400
Again, I think more options, uh, more modularity in being able to build, uh, the better for people involved.
00:27:06.480 --> 00:27:13.039
Although for me, I'm still gonna stick to my codex uh because it's just part of my everyday workflow.
00:27:13.359 --> 00:27:14.000
All right.
00:27:14.160 --> 00:27:19.920
Uh next, yes, this mod this model has been going absolutely viral.
00:27:20.079 --> 00:27:22.559
It is Minimax H3.
00:27:23.119 --> 00:27:25.920
So this is a new video model.
00:27:26.240 --> 00:27:42.720
Uh, and they did Minimax did just release the weights for H3, and it became the first open model to ever top a major AI video ranking, as it's number one in video editing on artificial analysis, and number two in text to video.
00:27:43.200 --> 00:27:48.319
So, artificial analysis called it the strongest open weights model by far.
00:27:48.480 --> 00:27:50.319
That's why this thing has been blowing up online.
00:27:50.559 --> 00:28:00.880
If you've seen any like really good AI video uh over the last like couple of days, like on Twitter or whatever, it's probably minimax, especially.
00:28:01.039 --> 00:28:09.680
And the reason why you're probably seeing it is because obviously, with this being uh in open weights Chinese model, there's like no copyright restrictions.
00:28:09.839 --> 00:28:27.599
I don't know why it's been like, at least on my feed, it's been like these 15-second Seinfeld clips, you know, of people talking about AI, or you know, 15-second clips from the office of people talking about AI, and they're obviously really good, right?
00:28:27.680 --> 00:28:31.119
Because it's just you know, ripping the copyright.
00:28:31.200 --> 00:28:38.240
It's you know, uh Michael Scott's face and Michael Scott's voice, but talking about you know, agentic swarms or something like that.
00:28:38.480 --> 00:28:41.039
So uh here's a little bit more about the model.
00:28:41.119 --> 00:28:53.039
So the model reads text, images, video, and audio as one context, and it outputs four to 15 second clips at up to 2k with native stereo audio.
00:28:53.200 --> 00:29:09.920
So that's if you're using their hosted service, but if you are uh using the uh actual um uh you know downloadable version, the the quantized version, you're not gonna be able to run the 2K uh resolution.
00:29:10.000 --> 00:29:15.359
It is gonna be a standard uh standard definition and it's gonna take a while, right?
00:29:15.519 --> 00:29:20.640
So let's just say you have like a you know 16 or 32 gigabyte MacBook.
00:29:20.720 --> 00:29:21.519
Let's just say that.
00:29:21.680 --> 00:29:29.119
Yeah, it's gonna take you maybe 20 or 30 minutes or longer um to make a 15-second clip.
00:29:29.279 --> 00:29:32.240
But the thing is, well, it's an option, right?
00:29:32.480 --> 00:29:37.680
Now we literally have models that are the state of the art, right?
00:29:37.759 --> 00:29:44.559
So, you know, number one, number two, and the two big categories, and you can download them and use them for free, right?
00:29:44.720 --> 00:29:54.880
So absolutely wild world where we're litting living in, and for the most part, you cannot tell that they're um AI videos, right?
00:29:55.119 --> 00:30:03.440
Uh like obviously you can't, because you know, if Jerry Seinfeld is losing his mind um over an agentic swarm, obviously, right?
00:30:03.680 --> 00:30:07.680
But I mean, everything else about it, it looks real, it sounds right.
00:30:07.759 --> 00:30:11.119
Jerry Seinfeld's voice, how he does his voice, he throws it all over the place.
00:30:11.200 --> 00:30:13.680
It gets it right, it nails it, right?
00:30:13.839 --> 00:30:30.400
Obviously, every once in a while you're gonna get some some bad generations, but the stuff I've seen online, I'm like, this is fairly impressive, but also to me, it's extremely scary because you know what takes like 15 seconds, yeah, like campaign videos, right?
00:30:30.559 --> 00:30:44.240
This is gonna absolutely wreck the upcoming US election because here's the thing, it's out in the wild, so it's not like any amount of uh you know new US policy is gonna stop this thing.
00:30:44.319 --> 00:30:51.440
So, yeah, I can see this thing causing a ton, a ton of trouble in a couple of months uh with the elections coming up.
00:30:51.680 --> 00:30:52.799
So, who has access?
00:30:52.960 --> 00:30:57.119
So, yeah, the hosted version, anyone anywhere, right, through the Minimax API.
00:30:57.200 --> 00:31:04.799
So, yeah, you don't need the fat you know$10,000 computer to you know eke out a 15-second video for that one.
00:31:04.880 --> 00:31:17.039
Uh, so you're paying the API prices, but the downloadable weights are on Hugging Face, but the license excludes right now uh the US, EU, UK, and South Korea from running it locally.
00:31:17.119 --> 00:31:20.480
So if you're in those regions, it says they're API only.
00:31:20.559 --> 00:31:26.160
It seems like most people are skirting around those um uh restrictions somehow.
00:31:26.640 --> 00:31:27.200
All right.
00:31:27.359 --> 00:31:28.720
Uh so why is it useful?
00:31:28.880 --> 00:31:31.200
Well, it's one model that can replace a lot.
00:31:31.279 --> 00:31:36.960
So it does text to video, image to video, first last frame, subject reference, motion reference.
00:31:37.119 --> 00:31:39.839
So it is much more than just a video generator.
00:31:40.000 --> 00:31:42.000
There's a lot that's going on under the hood.
00:31:42.160 --> 00:31:49.200
All right, last but not least, there is a new version of GPT 5.6 soul.
00:31:49.279 --> 00:31:55.680
So yes, open AI made a uh pretty notable update to its flagship model.
00:31:56.000 --> 00:31:57.599
Here's what's new.
00:31:57.920 --> 00:32:15.759
Uh, well, the big one is OpenAI made the free Chat GPT, the best of free AI has literally ever been, because now you have unlimited generations, even if you do not have a paid plan uh on GPT 5.6 Luna.
00:32:15.839 --> 00:32:25.440
So again, uh kind of the tiers, GPT 5.6 soul is the best, GPT 5.6 Terra is in the middle, and then you have GPT 5.6 Luna, still a very good capable model.
00:32:25.599 --> 00:32:32.079
Actually, GPT 5.6 Luna is pretty much on par uh with um Claude Sonnet 5.
00:32:32.160 --> 00:32:34.000
So just if you need a comparison.
00:32:34.559 --> 00:32:41.519
So uh for paid users, OpenAI merged the instant and thinking models into one.
00:32:41.839 --> 00:32:47.599
Uh so now you have a slider that controls how much reasoning it puts into each answer.
00:32:47.680 --> 00:32:54.720
So if you're on chatgbt.com, uh it's gonna seem a lot more familiar if you were using the desktop version.
00:32:54.880 --> 00:33:07.519
So, along with this, open AI claims 68% fewer responses containing factual errors versus GPT 5.6 instant on areas such as financial, medical, and legal prompts.
00:33:07.759 --> 00:33:10.640
So Zol just dropped its new.
00:33:10.799 --> 00:33:11.839
So who has access?
00:33:11.920 --> 00:33:15.440
Well, literally one billion users, yes.
00:33:15.599 --> 00:33:19.920
So uh one billion people have better models altogether.
00:33:20.079 --> 00:33:28.319
So this is not only free plans getting unlimited GPT 5.6 Luna, which I still don't understand how that's possible, right?
00:33:28.559 --> 00:33:32.160
How is it possible that, like, okay, let's just say anthropic, right?
00:33:32.240 --> 00:33:44.000
Let's just say for whatever reason, obviously they the you know, opus five um and fable five are like you know, one A, one B or one B, one C, depending on how you look at it, with GPT 5.6 soul.
00:33:44.160 --> 00:33:49.200
But after that, right, it's it's GPT 5, uh GPT or sorry, sonnet five, right?
00:33:49.920 --> 00:34:00.079
On Sonnet 5, you on a paid plan, right, in your five-hour window, you might get 10 to 20 prompts, right?
00:34:01.039 --> 00:34:02.160
On a paid plan.
00:34:02.480 --> 00:34:07.680
So now on a free plan on Chat GPT, you get unlimited.
00:34:08.000 --> 00:34:17.360
So this again, I like I know I sound like a broken record, and you know, there's plenty of stories uh, you know, um chronicling this online.
00:34:17.519 --> 00:34:20.400
It's just been a bad couple of weeks for anthropic, right?
00:34:20.480 --> 00:34:28.239
Between all these new open models coming out, Deep Seek comes out uh essentially cheap, like free, right?
00:34:28.320 --> 00:34:30.480
Uh OpenAI, we talked about this last week.
00:34:30.559 --> 00:34:34.639
They essentially made their GVD 5.6 Luna essentially free.
00:34:34.800 --> 00:34:38.559
Uh so yeah, it hasn't been a good couple of weeks for Anthropic.
00:34:38.719 --> 00:34:40.960
So uh we'll see where their number lands.
00:34:41.119 --> 00:34:44.559
They're expected to go public maybe in like as soon as two months.
00:34:44.639 --> 00:34:46.880
So that one uh should be interesting to see.
00:34:46.960 --> 00:34:49.119
So uh who's gonna find it valuable?
00:34:49.199 --> 00:34:51.440
Why is this useful from open AI?
00:34:51.599 --> 00:34:52.960
Well, uh a couple things.
00:34:53.039 --> 00:35:00.000
Well, there's no more having to choose between instant and thinking uh for paid users, it's just one model with a dial.
00:35:00.159 --> 00:35:04.800
And then obviously the unlimited free chat absolutely bonkers, right?
00:35:04.880 --> 00:35:21.440
This is, I think, open AI uh starting to in uh inch closer to you know being able to provide this level of intelligence to you know everyone in the world, which it seems like you know, that's what their goal is, you know, bringing AGI to the to the masses, right?
00:35:21.599 --> 00:35:23.840
Um, I still cannot get over.
00:35:24.480 --> 00:35:28.239
Like when I read that, when I saw this announcement, I'm like, this can't be right.
00:35:28.480 --> 00:35:34.480
Like, I'm like un unlimited GBD 5.6 because usually free users, right?
00:35:34.639 --> 00:35:39.679
Because it's all about compute, you know, usually free users are one tier below, right?
00:35:39.760 --> 00:35:43.440
Which is how it was, you know, yesterday, right?
00:35:43.679 --> 00:35:53.199
Uh you know, free users were on GBD 5.5, and that's how it's usually been, because free users make up the majority of that billion users.
00:35:53.360 --> 00:35:59.599
So um open AI had in in terms of post-training and inference, they've cracked something, right?
00:35:59.760 --> 00:36:02.639
I had kind of my uh recursive self-improvement show.
00:36:02.719 --> 00:36:10.239
It's like, yeah, we we no one has full RSI, but it seems like open AI's got something figured out.
00:36:10.480 --> 00:36:24.000
The fact that they lowered the price of GPD 5.6 Luna by 80%, and now they are giving a uh frontier level model uh away to a billion people unlimited, unlimited for free.
00:36:24.159 --> 00:36:24.800
Doesn't make sense.
00:36:24.880 --> 00:36:30.880
All right, anyways, I'll stop losing my marbles and let you get to go using all of these new AI features.
00:36:30.960 --> 00:36:31.840
I hope this was helpful.
00:36:31.920 --> 00:36:32.880
If so, let me know.
00:36:33.039 --> 00:36:38.559
Subscribe to the podcast on Apple and Spotify, then go to your everydayai.com.
00:36:38.719 --> 00:36:39.519
Thanks for tuning in.
00:36:39.599 --> 00:36:42.719
I'll see you Monday and next week for more everyday AI.
00:36:42.800 --> 00:36:43.760
Thanks, y'all.
00:00:01.040 --> 00:00:04.320
What if you could make a billion people smarter?
00:00:04.559 --> 00:00:07.280
Yeah, it's kind of what OpenAI just did.
00:00:07.440 --> 00:00:30.800
Uh, there's usually never been a free AI plan where I actually think it was good enough or reliable enough for everyday business use, but OpenAI just announced that as they quietly kind of announced that free users, so users on the free plan, get unlimited use of GPT 5.6 Luna, a really good and capable model.
00:00:31.039 --> 00:00:40.000
So previously, free Chat GPT users were usually served a combination of GPT 5.5 and GPT 5.4, depending on the thinking level they chose.
00:00:40.560 --> 00:00:45.439
Actually, free and low price AI really had a week in the limelight.
00:00:45.600 --> 00:00:57.280
So between that announcement from OpenAI, there was also a new open weight powerhouse model from Alibaba and a new model from Meta that's basically free if you let them train on your data.
00:00:57.600 --> 00:00:59.920
So you probably missed some of it.
00:01:00.000 --> 00:01:08.640
That's because between all the big model releases, there's actually dozens of smaller feature updates each week that can move the needle for your business.
00:01:08.799 --> 00:01:10.799
And if you miss one or two, it's no big deal.
00:01:10.879 --> 00:01:16.480
But if you really don't pay attention to the movements that happen between the big splashes, you'll miss out completely.
00:01:16.640 --> 00:01:24.640
And that's why on Fridays, we bring you the Friday features, our weekly update of the latest AI updates that you can actually use today.
00:01:24.879 --> 00:01:27.680
These aren't news or leaks or coming soons.
00:01:27.920 --> 00:01:31.840
These are just new AI features you can use this Friday.
00:01:32.000 --> 00:01:33.840
So let's get into it.
00:01:34.319 --> 00:01:36.959
So on today's show, here's what you're gonna learn.
00:01:37.040 --> 00:01:41.840
You're gonna learn how you can get essentially free frontier coding, or at least for free 99.
00:01:42.159 --> 00:01:55.760
If you let a certain company named Meta train on your data, you're gonna know what new open source standard there is for better agent orchestration and how basically all the big players in the game got in on it, except for one.
00:01:56.000 --> 00:02:02.640
And you're gonna know the new updates that give free Chat GPT users the best deal in AI right now.
00:02:02.799 --> 00:02:03.840
Let's jump into it.
00:02:04.000 --> 00:02:05.280
Welcome to Everyday AI.
00:02:05.439 --> 00:02:07.599
My name's Jordan Molson, and we do this every day.
00:02:07.760 --> 00:02:16.879
This is your daily live stream, podcast, and free daily newsletter, helping business leaders like you and me keep up with these AI updates because they quite literally don't stop.
00:02:16.960 --> 00:02:21.120
I tell you what's good, what's not, how to use it to grow your company and your career.
00:02:21.280 --> 00:02:30.000
So if that's exactly what you're trying to do, do me a favor, stranger, click that subscribe button on Apple Podcasts or Spotify, wherever you're listening.
00:02:30.080 --> 00:02:33.680
Uh, and then make sure you go to our website at your everydayai.com.
00:02:33.759 --> 00:02:34.479
It's in the show notes.
00:02:34.560 --> 00:02:37.759
So make sure you go check that out and sign up for our free daily newsletter.
00:02:37.840 --> 00:02:39.919
We're gonna be recapping the highlights from today's show.
00:02:40.000 --> 00:02:45.599
So if you miss anything, don't worry, it's in there as well as all the other AI updates you need to know to get ahead.
00:02:45.759 --> 00:02:49.280
All right, let's get into it and look live.
00:02:49.439 --> 00:02:59.120
All right, so yeah, if you are watching uh or sorry, if you're listening to this, there's always a video version, not too much that you need to see uh today, if I'm being honest.
00:02:59.360 --> 00:03:04.879
But um you can always check that out on our website at your everydayai.com.
00:03:05.039 --> 00:03:06.319
All right, here we go.
00:03:07.520 --> 00:03:09.199
Eventually, come on, screen.
00:03:09.280 --> 00:03:09.840
There we go.
00:03:09.919 --> 00:03:11.520
Yeah, it's unedited, folks.
00:03:11.680 --> 00:03:14.080
All right, our first big AI update.
00:03:14.240 --> 00:03:16.159
If you're creative, you're gonna like this one.
00:03:16.240 --> 00:03:22.639
That's because Adobe just launched a big update to its unified plugin in Chat GPT.
00:03:22.800 --> 00:03:32.400
Now it can access more than 70 tools from Photoshop, Firefly, Acrobat, Express, Premiere, Lightroom, Illustrator, InDesign, and Stock.
00:03:32.719 --> 00:03:34.400
So essentially, how does this work?
00:03:34.479 --> 00:03:39.039
Well, you describe what you want it to do, and it picks the right Adobe tools automatically.
00:03:39.199 --> 00:03:45.840
So you can edit photo sets, make campaign videos, uh, customize different design templates, turn spreadsheets into PDFs.
00:03:45.919 --> 00:03:51.520
So it kind of replaces the separate Adobe apps that were available in ChatGPT before.
00:03:51.599 --> 00:03:59.360
So now it's just one unified app where you don't even have to like know, oh, should I be using you know Acrobat for this?
00:03:59.520 --> 00:04:03.599
Should I be using uh, you know, Lightroom, Photoshop?
00:04:03.680 --> 00:04:04.400
It doesn't matter.
00:04:04.479 --> 00:04:06.080
So you just open Chat GPT.
00:04:06.159 --> 00:04:14.560
Uh, if you do have that Adobe Creative Crowd, uh Creative Cloud subscription, and you just dump in a bunch of files and say, do all this stuff to them, and it does it.
00:04:14.639 --> 00:04:19.680
Uh, so Adobe is running this play on every chat bot, not just Chat GPT.
00:04:19.759 --> 00:04:26.079
Uh, its tools are already on Claude and Copilot, but I believe this is the first um in terms of these types of features.
00:04:26.240 --> 00:04:28.879
This is the first one uh we've seen that's this robust.
00:04:29.040 --> 00:04:30.079
So who has access?
00:04:30.240 --> 00:04:34.160
Well, anyone, if you are on Chat GPT, you got it.
00:04:34.240 --> 00:04:39.839
And there's also four um mini uh features that are in included in this announcement.
00:04:40.000 --> 00:04:42.800
You don't need a plan uh to use them.
00:04:43.120 --> 00:04:47.439
So yeah, even right now, you don't need an Adobe account to start.
00:04:47.600 --> 00:05:02.240
Even just using guest mode does give you a broad set of tools immediately, not everything, because signing in with your Adobe account does unlock uh some of the more generative, uh, you know, heavy AI features, uh, and then being able to sync everything in your creative cloud.
00:05:02.319 --> 00:05:08.879
So it also does work inside of ChatGPT work in codecs so it's not just on the uh ChatGPT.com release.
00:05:09.120 --> 00:05:10.879
So who's this useful?
00:05:11.040 --> 00:05:11.920
Well, anyone.
00:05:12.000 --> 00:05:18.160
If you're trying to do something creative, if you are a creative, if you're not creative, but you want to be creative, this is a great way to do it.
00:05:18.240 --> 00:05:32.800
So obviously, open AI and ChatGPT has a fantastic, literally state of the art um kind of their uh GPT images too, which is really, really good, but it can do everything that you might want it to.
00:05:32.959 --> 00:05:33.600
Um, right.
00:05:33.680 --> 00:05:38.240
So in those cases, now you have this tool from uh Adobe.
00:05:38.399 --> 00:05:40.319
All right, so that's a good one.
00:05:40.480 --> 00:05:48.639
But here's uh maybe even a better one, maybe for a lot of our audience, because meta, another pretty big splash here.
00:05:48.800 --> 00:05:58.800
And there's actually a plan here that I wasn't expecting from meta, and I don't hate it as one that I'm just like, hey company, take my data.
00:05:59.040 --> 00:06:06.079
I don't hate this new pricing from Meta Muse, Spark, and their new Muse Code and Spark 1.2.
00:06:06.160 --> 00:06:11.759
So here's what's new Meta released Muse Code, which is their kind of take on Claude Code or Codec.
00:06:11.920 --> 00:06:17.920
So that's a terminal coding agent right now, uh, powered by its new Muse Spark 1.2 model.
00:06:18.000 --> 00:06:19.360
So, yeah, two different things.
00:06:19.439 --> 00:06:23.199
Uh, it is Muse Code, so that's a terminal coding agent, right?
00:06:23.279 --> 00:06:35.439
So, not like the desktop version of Cloud Code or Codecs, like the CLI version of Cloud Code or Codex, and then uh the next one is the new model, uh Muse Spark 1.2.
00:06:36.000 --> 00:06:42.319
So the pricing here, even if you're paying the full API pricing, pretty good rates in terms of performance.
00:06:42.480 --> 00:06:48.720
It's a dollar 25 per million tokens in and four dollars and 25 cents per million tokens out.
00:06:48.959 --> 00:06:54.879
But there is a new what they're calling contributor tier, it's an interesting name for it.
00:06:54.959 --> 00:07:04.240
Um, that drops that price down to 10 cents in and 20 cents out if you let Meta train on your data.
00:07:04.399 --> 00:07:10.959
So that's roughly 95% off in exchange for essentially your code, right?
00:07:11.199 --> 00:07:21.759
So um at first I'm like, absolutely no one's gonna do this, but then I'm like, yeah, like I don't know, especially if you're starting from scratch, uh, right?
00:07:21.839 --> 00:07:30.720
Let's just say that you have a personal project, or you know, if you're an indie hacker, indie developer, I'm like, that's actually not a terrible idea, right?
00:07:30.879 --> 00:07:36.879
Obviously, if you're working with large proprietary anything enterprise, no one would ever touch that contributor tier.
00:07:36.959 --> 00:07:43.680
But I'm like, for everyone else, that it's more of a hobbyist, it's actually, I think, not that bad of a deal.
00:07:44.160 --> 00:07:50.720
So um, every model call, tool run, approval, and edit uh does get written to a local event log.
00:07:50.800 --> 00:07:54.319
So if the agent crashes, it can resume where it stopped.
00:07:54.480 --> 00:07:55.920
So who has access to this?
00:07:56.000 --> 00:07:58.000
Well, it's open now in beta.
00:07:58.240 --> 00:08:01.759
You can literally just go run a terminal command and you're off and running.
00:08:01.920 --> 00:08:17.519
Uh so the new model though, meta spark uh sorry, Muse Spark 1.2, is available inside Muse Code and to any developer through the meta uh meta model API, and it does have a 1 million token context window.
00:08:17.920 --> 00:08:20.079
So uh why is this useful?
00:08:20.240 --> 00:08:30.480
Well, aside from writing code, if you need persistent background sub agents that can work on hard tasks on their own and decide when to report back, uh it's pretty good for that.
00:08:30.639 --> 00:08:37.759
Uh, you know, the scores so far um for what's available on artificial analysis as an example, pretty good, right?
00:08:38.000 --> 00:08:46.639
You know, all of a sudden, meta went from not very relevant for like a full year and a half to now all of a sudden they're in the conversation, right?
00:08:46.799 --> 00:08:51.759
They're again Google might change this, you know, any any minute now, right?
00:08:51.840 --> 00:08:52.559
Watch Google.
00:08:52.639 --> 00:09:02.480
There's been uh rumors that we're getting something, you know, maybe this week or August 10th, uh, either Gemini 3.5 Pro or if they skip straight to Gemini 4 Pro, who knows?
00:09:02.639 --> 00:09:06.159
But Meta is actually ahead of Google, right?
00:09:06.240 --> 00:09:10.080
Not just in terms of model capabilities, but on pricing as well.
00:09:10.240 --> 00:09:15.200
So Meta has had a very impressive uh last month or two.
00:09:15.360 --> 00:09:33.919
So I think if you're you know a dev working on a large uh repo, if you want long running agents, like I said, or if you're an indie dev or a cost-sensitive team and you're not necessarily working on you know proprietary data, uh that contributor tier would just still a funny name for it.
00:09:34.080 --> 00:09:35.039
Not bad, right?
00:09:35.200 --> 00:09:40.799
Uh, we'll see if Muse code uh goes for very long, right?
00:09:41.039 --> 00:09:44.559
Um, I don't know, maybe this is the non-developer in me speaking.
00:09:44.639 --> 00:10:03.039
I've always been able to do the basics in a terminal, but I feel even the people that have been hardcore, like, hey, I'm a terminal coder, you know, in in terms of being a software engineer or dev, I've seen even some of the people that, you know, a year ago were like, I'm never giving up the terminal to being like the terminal just doesn't make sense anymore.
00:10:03.200 --> 00:10:03.919
So I don't know.
00:10:04.080 --> 00:10:04.559
We'll see.
00:10:04.639 --> 00:10:16.480
I mean, you have to start somewhere with uh, you know, in terms of having a coding agent, but regardless, meta's showing where they're gonna play, they're gonna try to compete with anthropic in Google on token pricing.
00:10:16.559 --> 00:10:18.960
I don't think that's necessarily OpenAI's game.
00:10:19.039 --> 00:10:22.879
Uh, that is where uh Anthropic reportedly gets about 80% of its revenue.
00:10:22.960 --> 00:10:32.000
So this to me looks like a shot at Anthropic, um, you know, bringing the actual coding agent with views code, uh, extremely cheap and affordable.
00:10:32.240 --> 00:10:34.240
And the performance is there.
00:10:34.879 --> 00:10:35.360
All right.
00:10:35.679 --> 00:10:37.200
Our next piece of AI news.
00:10:37.360 --> 00:10:40.720
Speaking of, well, very affordable uh models.
00:10:40.879 --> 00:10:45.919
Well, we have a new uh not quite king of the open hill.
00:10:46.000 --> 00:10:55.519
It's kind of uh actually let me check that because I know artificial analysis just put out a new version um of their four point, what is it, four point one one now?
00:10:55.600 --> 00:10:58.720
So I'm I'm double checking why to see uh okay, yeah.
00:10:58.879 --> 00:11:04.879
So they are just a little bit behind Kimi K3, uh, but essentially uh it is in the same tier.
00:11:04.960 --> 00:11:10.000
So Alibaba's new uh open weights Quinn 3.8 Max.
00:11:10.080 --> 00:11:11.840
Um, it is in the same tier.
00:11:11.919 --> 00:11:20.879
So it is in that uh, you know, GPT 5.6, uh Claude Fable 5, Claude Opus 5, Kimmy K3, and Quen 3.8.
00:11:21.120 --> 00:11:23.840
That's kind of your top five models in the world now.
00:11:23.919 --> 00:11:29.120
You know, we have two open weights models from China uh in spots four and five.
00:11:29.279 --> 00:11:38.799
So here is what it is it is Quen 3.8 Max, a 2.4 trillion parameter open weight model with a million token context window.
00:11:38.879 --> 00:11:41.120
It takes text, images, and video.
00:11:42.080 --> 00:11:46.320
Um, so here's the um interesting thing.
00:11:46.480 --> 00:11:51.440
So it did become uh the top-ranked Chinese model for text only.
00:11:51.600 --> 00:11:53.759
All right, so for text only, it takes that.
00:11:54.000 --> 00:12:11.279
And here's the thing, though, there was a report literally hours ago uh from Reuters that said Alibaba plans to begin charging major enterprise users for Quen 3.8 Max to charge them a share of their revenue they generate starting next week.
00:12:11.440 --> 00:12:13.039
So that one's interesting.
00:12:13.200 --> 00:12:15.120
So who has access right now?
00:12:15.200 --> 00:12:16.320
Uh well, anyone.
00:12:16.399 --> 00:12:27.200
So if you're working on the hosted API or through Alibaba Cloud Model Studio, uh so yeah, if you are in Alibaba Cloud Accounts, you can use it right now.
00:12:27.600 --> 00:12:30.639
Uh and the API is OpenAI compatible.
00:12:30.720 --> 00:12:34.960
So switching is a base URL and model uh ID change.
00:12:35.039 --> 00:12:40.480
So it's open weights, although the weights are not out yet, uh, but they did promise them within days of the release.
00:12:40.559 --> 00:12:42.559
That's essentially what Kimmy did as well.
00:12:42.639 --> 00:12:46.000
Um, and they kind of hit their target deadline for releasing the weights.
00:12:46.159 --> 00:12:50.960
But when the weights land, obviously you can't like self-host this.
00:12:51.039 --> 00:12:51.759
Um, right.
00:12:51.919 --> 00:13:01.120
And again, unless you happen to have a data center, uh, you know, a bunch of GPUs that you just weren't using in your basement, you aren't gonna use this, right?
00:13:01.200 --> 00:13:08.480
It's kind of this um uh you know, I don't know, fortunate, unfortunate uh current dichotomy of open weight open source software, right?
00:13:08.639 --> 00:13:12.080
I think a year ago, the best open models you could, you know, download.
00:13:12.159 --> 00:13:14.879
If you had a big expensive Mac, you could do it.
00:13:14.960 --> 00:13:16.000
You can't anymore.
00:13:16.080 --> 00:13:26.960
Uh, right, these multiple trillion parameter uh you know open weight models, they're not really for running locally anymore unless you literally have a data center.
00:13:27.440 --> 00:13:29.120
So, well, why is this useful?
00:13:29.200 --> 00:13:32.720
Well, it's frontier level capability with open weights.
00:13:32.879 --> 00:13:39.600
Um, so yeah, if for bigger companies, you can, you know, fine-tune these models, or if you're working with a provider, you can fine-tune them.
00:13:39.679 --> 00:13:44.639
So, unlike proprietary models, uh, where it's not always as easy to do something like that.
00:13:44.879 --> 00:13:53.360
Um, so uh also interesting to note, it did debut number four on the front-end code arena.
00:13:53.600 --> 00:13:59.440
Uh, so you know, apparently it has some pretty good front-end uh taste to it as well.
00:13:59.519 --> 00:14:01.200
So, who's gonna find this valuable?
00:14:01.360 --> 00:14:04.080
I think really just large enterprises.
00:14:04.240 --> 00:14:04.639
That's it.
00:14:04.799 --> 00:14:16.960
So if you're hedging against, you know, US lab pricing and lock-in, uh, or just you know, people who are already self-hosting large models, researchers, uh, they are coming out with a 27B version.
00:14:17.200 --> 00:14:25.120
Uh, you know, so if you do have uh, you know, a couple heavy Mac studios beefed together, you can run it like super slow or something like that.
00:14:25.279 --> 00:14:31.120
Uh actually 27B, yeah, you should be able to run it on like the most expensive DGXs.
00:14:31.279 --> 00:14:35.759
Um, but yeah, again, nice open model, uh, open weight model.
00:14:35.919 --> 00:14:43.200
So if nothing else, uh it's pushing the other open models, which are pushing proprietary models, which are push pushing prices down, right?
00:14:43.279 --> 00:14:45.519
That's ultimately what's happening here.
00:14:45.600 --> 00:14:58.639
Uh, because the people who can afford uh to run something like a Quen uh Quen 3.8 Max by running it locally are those, you know, Fortune 500 types that prep that may have access to compute.
00:14:58.799 --> 00:15:06.639
So that is the people who are generally, you know, using anthropics models would probably be their first customer, OpenAI, Google maybe as well.
00:15:06.799 --> 00:15:14.000
So, you know, that's also going to push uh in theory, you know, either push down the prices from the major frontier labs or at least keep them in check.
00:15:14.080 --> 00:15:18.879
So even if you're like, okay, this model isn't gonna impact me or my business, we can't run it.
00:15:19.039 --> 00:15:20.159
Well, you can, right?
00:15:20.240 --> 00:15:22.639
There's always you know providers that you can use.
00:15:22.799 --> 00:15:26.799
Um, but if anything else, keep models in check.
00:15:26.960 --> 00:15:28.559
All right, let's go next.
00:15:28.720 --> 00:15:34.399
Uh, next, we already covered this one in depth on our Wednesday AI Working Wednesday show.
00:15:34.480 --> 00:15:36.320
That's episode 834.
00:15:36.720 --> 00:15:40.000
But a ton new with Gemini Notebook.
00:15:40.080 --> 00:15:48.639
Yes, the tool formerly known as Notebook LM finally got its big glow up that rolled out to all paid users.
00:15:48.799 --> 00:15:54.080
So this is technically an older update, but we didn't cover it on the show previously.
00:15:54.159 --> 00:15:58.480
Well, aside from Wednesday, uh, when it was, you know, I think it was just released on Tuesday.
00:15:58.639 --> 00:16:02.320
We didn't really cover it before because it was only available to ultra subscribers.
00:16:02.480 --> 00:16:09.600
So just so you know, on these Friday shows, these are just you know tools or updates that are available to your standard, you know,$20 base plans.
00:16:09.679 --> 00:16:14.320
But now we have that Gemini Agenc notebook that is out.
00:16:14.480 --> 00:16:16.000
So yeah, has a new name.
00:16:16.240 --> 00:16:22.879
Notebook LM is gone, Gemini Notebook is here, and Gemini Notebook, at least for paid users, is agentic by default.
00:16:23.039 --> 00:16:29.120
So Google finished rolling out the agentic version of Gemini Notebook, the upgrade to pro subscribers.
00:16:29.279 --> 00:16:39.919
So every notebook now gets a secure cloud computer that writes and runs real code against your sources, plus auto-generated charts, spreadsheets, slide decks, and PDFs.
00:16:40.080 --> 00:16:47.840
So it runs on Gemini 3.5, but it also uses the anti-gravity harness and it has a hundred built-in skills.
00:16:48.080 --> 00:16:52.080
So think of like code interpreter grounded in your documents.
00:16:52.240 --> 00:16:59.919
Uh, you can also just, you know, start a blank notebook, give it a vague query, and it can also now use the web, although it will confirm with you, right?
00:17:00.000 --> 00:17:05.519
Because that's the good thing about, you know, previously Notebook LM is it stayed completely grounded in your source data.
00:17:05.680 --> 00:17:12.240
So now uh Gemini Notebook can query the web, uh, which is a little bit different for me, right?
00:17:12.400 --> 00:17:19.359
But it will at least let you know hey, you know, if you're asking about something that's not in your source material, it will tell you I don't have the answer.
00:17:19.519 --> 00:17:21.440
Do you want me to go search the web?
00:17:22.240 --> 00:17:24.720
Uh so uh who has access?
00:17:24.960 --> 00:17:28.559
Anyone with a uh Gemini Pro account?
00:17:28.799 --> 00:17:30.400
Uh, you've got it now.
00:17:30.480 --> 00:17:32.799
Uh so this is on the web version only.
00:17:32.960 --> 00:17:35.680
If you're an ultra subscriber, you already have this.
00:17:35.839 --> 00:17:40.240
Uh free users, you're still gonna see Gemini Notebook, right?
00:17:40.319 --> 00:17:45.599
Uh, but you're not gonna have these new agenc the code execution or some of these new output formats.
00:17:45.680 --> 00:17:46.960
So that's the other thing that's new.
00:17:47.039 --> 00:17:53.599
Uh in Gemini Notebook, you can output as a PDF, uh, a doc, um, you know, different PNGs, uh, right.
00:17:53.680 --> 00:17:58.160
So essentially before you were a little more limited in what you could actually output.
00:17:58.240 --> 00:18:02.880
Uh, there was, you know, I think about eight different sources on the right hand side in the studio panel.
00:18:03.039 --> 00:18:05.839
Now you can just say, you know, output in these different sources.
00:18:05.920 --> 00:18:09.519
So I actually did what I thought was a pretty good rundown on Wednesday.
00:18:09.599 --> 00:18:12.079
So if you do want to check that out, make sure you go do that.
00:18:12.240 --> 00:18:13.680
Episode 834.
00:18:13.839 --> 00:18:15.359
Who's gonna find this valuable?
00:18:15.599 --> 00:18:17.440
Literally anyone, right?
00:18:17.599 --> 00:18:24.720
Um, if you're you know a heavy uh chat GPT work, codex, claud code, uh user, right?
00:18:24.799 --> 00:18:28.559
As an example, these new agentic features are pretty good, right?
00:18:28.720 --> 00:18:42.960
Um, I'm really looking forward to when um you know Google does introduce whether it's Gemini 3.5 or Gemini 4 Pro, whichever one, and then seeing those roll out to Notebook LM, I'll be really excited for because they're technically running the Gemini 3.
00:18:46.160 --> 00:18:50.000
But the Gemini 3.5 flash model is not that great.
00:18:50.160 --> 00:18:54.079
Um, so yeah, it's got like a 30.
00:18:54.799 --> 00:18:55.359
What is it?
00:18:55.519 --> 00:18:58.319
Mid no, that's Gemini 3.6 flash.
00:18:58.400 --> 00:18:59.039
Come on, come on.
00:18:59.119 --> 00:19:01.279
Where's where's my Gemini 3 flash?
00:19:01.440 --> 00:19:03.440
Uh let's see, let's see where we have it.
00:19:03.599 --> 00:19:06.640
It is okay, a 52 on the artificial analysis.
00:19:06.799 --> 00:19:08.960
Uh, so not that good.
00:19:09.119 --> 00:19:15.680
So it is below, um, or no, sorry, it is at the same level as uh GPT-5.6 Luna.
00:19:15.759 --> 00:19:18.960
So okay, from that respect, it's okay-ish, right?
00:19:19.039 --> 00:19:25.519
So if you think that open AI has three tiers, um, you know, they have uh Seoul, Terra, and Luna.
00:19:25.680 --> 00:19:32.480
Right now, Google's best tier is the lowest tier for um OpenAI's Chat GPT.
00:19:32.559 --> 00:19:36.000
So take that for uh you know what it's worth.
00:19:36.160 --> 00:19:41.680
So not the best model in the world um that's powering it, especially when you're a paid user, right?
00:19:41.759 --> 00:19:48.640
We're gonna talk about the free uh kind of upgrade for uh uh GBT56 Luna, which I think is great for free users, right?
00:19:48.720 --> 00:20:01.440
But for being a paid user right now, it's kind of not a great time uh to be a you know Gemini first company, but I will expect Google any day, any minute, any week now to come out with a good 3.5 pro or Gemini 4 Pro.
00:20:01.680 --> 00:20:04.160
And I hope it gets rolled out to Gemini Notebook.
00:20:04.319 --> 00:20:10.240
All right, next we have a new open standard for agent plugins, and this is pretty cool.
00:20:10.400 --> 00:20:13.279
So here's what's new, and yes, it's available now.
00:20:13.359 --> 00:20:19.839
There's a new open vendor neutral standard for packaging AI agent extensions.
00:20:20.000 --> 00:20:26.960
So now it's a one plugin, uh folder with a plug-in JSON file bundling and agent skill and mcp servers.
00:20:27.119 --> 00:20:35.839
And right now it works across ChatGPT, uh, codecs, cursor, Google, GitHub, Copilot, uh, VS Code, and others.
00:20:35.920 --> 00:20:41.200
So essentially, you can build an agent plugin once and run it anywhere.
00:20:41.839 --> 00:20:45.680
So this is a little confusing, but let me just break it down.
00:20:45.759 --> 00:20:53.279
Uh, and hopefully this is like 99% correct because I haven't actually gone out and built this because it's only an hour like in a couple hours old.
00:20:53.440 --> 00:20:56.559
Um, so previously think you have skills, right?
00:20:56.640 --> 00:21:00.720
And skills are essentially just a bunch of text files in a folder, uh, right.
00:21:00.880 --> 00:21:09.599
And then you also have things like plugins, and plugins are you know multiple skills or a skill in an app together, um, right?
00:21:09.759 --> 00:21:12.240
And then you have like JSON files.
00:21:12.400 --> 00:21:15.920
Um, so it's kind of like the new agent plugins.
00:21:16.000 --> 00:21:20.079
Um, you know, and then you know, you have your skills are in markdown files, right?
00:21:20.160 --> 00:21:24.640
It's essentially pulling all of these things together, but in an open standard, right?
00:21:24.720 --> 00:21:27.519
So that's great because you had the skills MD.
00:21:27.680 --> 00:21:34.799
Uh you know, skills are pretty universal, um, right, because they are just text files, they're markdown files.
00:21:34.880 --> 00:21:36.799
Um, so those are pretty universal.
00:21:36.960 --> 00:21:41.759
Uh, also inside the uh agent plugins, sorry, I forgot to mention this, it's MCPs as well.
00:21:41.920 --> 00:21:51.279
So it's essentially stacking, you know, your skills, your MCP servers, um, you, you know, your JSON files, uh, you know, all in one package.
00:21:51.440 --> 00:21:58.559
So this is nice uh because before you would have to load all of these things in individually, uh, which takes a lot of time.
00:21:58.720 --> 00:22:03.599
And when you you know it's Slightly different across different providers.
00:22:04.160 --> 00:22:09.039
You don't always have that from a business strategy perspective.
00:22:09.200 --> 00:22:12.079
You get a little locked in at times, which you don't want.
00:22:12.240 --> 00:22:18.960
So this is why it was great to see almost every single big name instantly support the Asian plugins, right?
00:22:19.119 --> 00:22:21.359
Anthropic, obviously not on the list.
00:22:21.839 --> 00:22:25.839
I don't know why Anthropic just won't get on board with this.
00:22:26.079 --> 00:22:30.000
Um, you know, obviously they have their Claude.md, right?
00:22:30.160 --> 00:22:36.720
But it's like, okay, Anthropic, just change it to Agent MD, like literally every other company, right?
00:22:36.880 --> 00:22:40.799
Google, Microsoft, Cursor, OpenAI, everyone else is doing this.
00:22:40.960 --> 00:22:42.160
Anthropic, get on board.
00:22:42.319 --> 00:22:46.720
It's just making uh it's I think it's slowing the pace of AI down.
00:22:46.799 --> 00:22:49.680
Uh, I think it's a maybe a bad look on anthropic.
00:22:49.759 --> 00:22:52.480
Like, are you just trying to lock people in, right?
00:22:52.559 --> 00:22:55.200
Maybe I don't know, maybe they are, maybe that's part of their business strategy.
00:22:55.359 --> 00:23:03.599
But for me, right, I love uh kind of having this open, uh, you know, new agent plug-in standard because it does make it easier for businesses.
00:23:03.759 --> 00:23:13.119
Uh, the like the other thing, too, is um, yes, in most cases, I think it's best for companies to kind of pick one AI kind of operating system and stick there.
00:23:13.279 --> 00:23:17.680
But you know, you might maybe your finance team needs a different one, right?
00:23:17.839 --> 00:23:27.759
Maybe 90% of your company can be on uh you know Chat GPT work, but maybe 10% of your company needs to be on Gemini for whatever reason, right?
00:23:27.839 --> 00:23:34.799
And in this case, when you have these ancient plugins, it's very easy uh to have that standard and to be able to transfer that over.
00:23:34.960 --> 00:23:40.000
But you know, if anthropic doesn't get on board, it's not gonna be as easy for them, anyways.
00:23:40.319 --> 00:23:43.440
So openAI announced this, but it's not open AI's.
00:23:43.519 --> 00:23:51.920
So Vercell initiated it, uh, as well as uh Amazon, any sphere, which is cursor, GitHub Copilot, Microsoft, uh, etc.
00:23:52.240 --> 00:23:54.960
So um who has access right now?
00:23:55.039 --> 00:23:55.759
Well, anyone.
00:23:55.839 --> 00:23:57.519
So you can just build this, right?
00:23:57.599 --> 00:24:00.559
You can just point your coding agent to this uh the website.
00:24:00.640 --> 00:24:04.799
The website is just agent-plugins.org, right?
00:24:04.960 --> 00:24:06.400
All of the documentation is there.
00:24:06.480 --> 00:24:12.240
So this is just an easy way, especially if your team is becoming more agentic in your work, right?
00:24:12.319 --> 00:24:17.599
You have multiple MCP files, you have multiple skills, you have JSON files holding it all together.
00:24:17.839 --> 00:24:21.680
It's just an easier way to port this across server to server.
00:24:21.839 --> 00:24:23.759
All right, or I'm sorry, service to service.
00:24:23.920 --> 00:24:24.559
All right, next one.
00:24:24.720 --> 00:24:26.160
Speaking of being open, right?
00:24:26.240 --> 00:24:28.880
It's a big week for open and free.
00:24:29.039 --> 00:24:31.599
Uh, this one was not expecting this.
00:24:31.759 --> 00:24:36.640
Cloudflare announced their open source Cloudflare OS.
00:24:36.960 --> 00:24:41.839
Uh, it's an internal AI workspace that it built to actually run its own company.
00:24:41.920 --> 00:24:45.920
So they essentially said, we build this to run our company, so everyone else can have it as well.
00:24:46.079 --> 00:25:00.319
So uh essentially, every employee gets a browser-based HHID workspace with governed access to internal systems, research documents tied to live data, automated workflows, and shareable internal apps with well, no developers really need it.
00:25:00.400 --> 00:25:07.119
Uh, so thousands of Cloudflare's own employees across every team already use it daily, according to the company.
00:25:07.279 --> 00:25:11.039
So seems pretty battle-tested, not just like some demo software.
00:25:11.200 --> 00:25:24.799
So it competes technically, right, with Chad GPT Work or Cloud Co-Work or Gemini Enterprise, but it's free, open source, and runs in your own Cloudflare account instead of the kind of per seat SaaS.
00:25:25.279 --> 00:25:28.000
So uh who has access right now?
00:25:28.240 --> 00:25:30.799
Well, it's open source, so anyone.
00:25:30.880 --> 00:25:32.880
So you can go download the repo.
00:25:32.960 --> 00:25:41.920
Uh, you do obviously need a Cloudflare account to run it because it deploys into your own account uh using your own access policies and integrations.
00:25:42.000 --> 00:25:46.880
Uh, but your IT team, uh, essentially, yeah, they'll probably need to help you a little bit.
00:25:47.039 --> 00:25:56.240
Uh, but there is also a managed version through Cloudflare plus implementation partners uh that um are coming, but no date is given yet.
00:25:56.480 --> 00:25:59.839
So uh why is this useful?
00:26:00.000 --> 00:26:07.680
Well, you can just essentially bring your own model um and have an open source agentic platform to work on.
00:26:07.839 --> 00:26:14.720
So, yes, uh, this isn't free uh to use, it's just a free operating system.
00:26:14.799 --> 00:26:24.319
Think of it that way, but you still have to bring your apps, aka, you still have to bring your own model through the Cloudflare AI model gateway.
00:26:24.480 --> 00:26:28.640
Um, so but there's you know, you can do spend caps and you're not locked in.
00:26:28.720 --> 00:26:29.920
So think of it more like that.
00:26:30.000 --> 00:26:33.200
Think of it, you know, and there are other versions of this, right?
00:26:33.279 --> 00:26:40.400
Uh, there are versions of open, you know, open versions of um, you know, desktop agents, right?
00:26:40.480 --> 00:26:45.839
But this is from Cloudflare, one of the literally biggest names uh in the internet ever.
00:26:46.000 --> 00:26:48.160
Uh, so a very trusted source.
00:26:48.319 --> 00:26:54.079
So uh, you know, a lot of companies run their uh you know web operations through Cloudflare.
00:26:54.319 --> 00:26:55.920
So trusted, trusted name.
00:26:56.000 --> 00:26:59.519
Uh love this because it is giving people more options.
00:26:59.680 --> 00:27:06.400
Again, I think more options, uh, more modularity in being able to build, uh, the better for people involved.
00:27:06.480 --> 00:27:13.039
Although for me, I'm still gonna stick to my codex uh because it's just part of my everyday workflow.
00:27:13.359 --> 00:27:14.000
All right.
00:27:14.160 --> 00:27:19.920
Uh next, yes, this mod this model has been going absolutely viral.
00:27:20.079 --> 00:27:22.559
It is Minimax H3.
00:27:23.119 --> 00:27:25.920
So this is a new video model.
00:27:26.240 --> 00:27:42.720
Uh, and they did Minimax did just release the weights for H3, and it became the first open model to ever top a major AI video ranking, as it's number one in video editing on artificial analysis, and number two in text to video.
00:27:43.200 --> 00:27:48.319
So, artificial analysis called it the strongest open weights model by far.
00:27:48.480 --> 00:27:50.319
That's why this thing has been blowing up online.
00:27:50.559 --> 00:28:00.880
If you've seen any like really good AI video uh over the last like couple of days, like on Twitter or whatever, it's probably minimax, especially.
00:28:01.039 --> 00:28:09.680
And the reason why you're probably seeing it is because obviously, with this being uh in open weights Chinese model, there's like no copyright restrictions.
00:28:09.839 --> 00:28:27.599
I don't know why it's been like, at least on my feed, it's been like these 15-second Seinfeld clips, you know, of people talking about AI, or you know, 15-second clips from the office of people talking about AI, and they're obviously really good, right?
00:28:27.680 --> 00:28:31.119
Because it's just you know, ripping the copyright.
00:28:31.200 --> 00:28:38.240
It's you know, uh Michael Scott's face and Michael Scott's voice, but talking about you know, agentic swarms or something like that.
00:28:38.480 --> 00:28:41.039
So uh here's a little bit more about the model.
00:28:41.119 --> 00:28:53.039
So the model reads text, images, video, and audio as one context, and it outputs four to 15 second clips at up to 2k with native stereo audio.
00:28:53.200 --> 00:29:09.920
So that's if you're using their hosted service, but if you are uh using the uh actual um uh you know downloadable version, the the quantized version, you're not gonna be able to run the 2K uh resolution.
00:29:10.000 --> 00:29:15.359
It is gonna be a standard uh standard definition and it's gonna take a while, right?
00:29:15.519 --> 00:29:20.640
So let's just say you have like a you know 16 or 32 gigabyte MacBook.
00:29:20.720 --> 00:29:21.519
Let's just say that.
00:29:21.680 --> 00:29:29.119
Yeah, it's gonna take you maybe 20 or 30 minutes or longer um to make a 15-second clip.
00:29:29.279 --> 00:29:32.240
But the thing is, well, it's an option, right?
00:29:32.480 --> 00:29:37.680
Now we literally have models that are the state of the art, right?
00:29:37.759 --> 00:29:44.559
So, you know, number one, number two, and the two big categories, and you can download them and use them for free, right?
00:29:44.720 --> 00:29:54.880
So absolutely wild world where we're litting living in, and for the most part, you cannot tell that they're um AI videos, right?
00:29:55.119 --> 00:30:03.440
Uh like obviously you can't, because you know, if Jerry Seinfeld is losing his mind um over an agentic swarm, obviously, right?
00:30:03.680 --> 00:30:07.680
But I mean, everything else about it, it looks real, it sounds right.
00:30:07.759 --> 00:30:11.119
Jerry Seinfeld's voice, how he does his voice, he throws it all over the place.
00:30:11.200 --> 00:30:13.680
It gets it right, it nails it, right?
00:30:13.839 --> 00:30:30.400
Obviously, every once in a while you're gonna get some some bad generations, but the stuff I've seen online, I'm like, this is fairly impressive, but also to me, it's extremely scary because you know what takes like 15 seconds, yeah, like campaign videos, right?
00:30:30.559 --> 00:30:44.240
This is gonna absolutely wreck the upcoming US election because here's the thing, it's out in the wild, so it's not like any amount of uh you know new US policy is gonna stop this thing.
00:30:44.319 --> 00:30:51.440
So, yeah, I can see this thing causing a ton, a ton of trouble in a couple of months uh with the elections coming up.
00:30:51.680 --> 00:30:52.799
So, who has access?
00:30:52.960 --> 00:30:57.119
So, yeah, the hosted version, anyone anywhere, right, through the Minimax API.
00:30:57.200 --> 00:31:04.799
So, yeah, you don't need the fat you know$10,000 computer to you know eke out a 15-second video for that one.
00:31:04.880 --> 00:31:17.039
Uh, so you're paying the API prices, but the downloadable weights are on Hugging Face, but the license excludes right now uh the US, EU, UK, and South Korea from running it locally.
00:31:17.119 --> 00:31:20.480
So if you're in those regions, it says they're API only.
00:31:20.559 --> 00:31:26.160
It seems like most people are skirting around those um uh restrictions somehow.
00:31:26.640 --> 00:31:27.200
All right.
00:31:27.359 --> 00:31:28.720
Uh so why is it useful?
00:31:28.880 --> 00:31:31.200
Well, it's one model that can replace a lot.
00:31:31.279 --> 00:31:36.960
So it does text to video, image to video, first last frame, subject reference, motion reference.
00:31:37.119 --> 00:31:39.839
So it is much more than just a video generator.
00:31:40.000 --> 00:31:42.000
There's a lot that's going on under the hood.
00:31:42.160 --> 00:31:49.200
All right, last but not least, there is a new version of GPT 5.6 soul.
00:31:49.279 --> 00:31:55.680
So yes, open AI made a uh pretty notable update to its flagship model.
00:31:56.000 --> 00:31:57.599
Here's what's new.
00:31:57.920 --> 00:32:15.759
Uh, well, the big one is OpenAI made the free Chat GPT, the best of free AI has literally ever been, because now you have unlimited generations, even if you do not have a paid plan uh on GPT 5.6 Luna.
00:32:15.839 --> 00:32:25.440
So again, uh kind of the tiers, GPT 5.6 soul is the best, GPT 5.6 Terra is in the middle, and then you have GPT 5.6 Luna, still a very good capable model.
00:32:25.599 --> 00:32:32.079
Actually, GPT 5.6 Luna is pretty much on par uh with um Claude Sonnet 5.
00:32:32.160 --> 00:32:34.000
So just if you need a comparison.
00:32:34.559 --> 00:32:41.519
So uh for paid users, OpenAI merged the instant and thinking models into one.
00:32:41.839 --> 00:32:47.599
Uh so now you have a slider that controls how much reasoning it puts into each answer.
00:32:47.680 --> 00:32:54.720
So if you're on chatgbt.com, uh it's gonna seem a lot more familiar if you were using the desktop version.
00:32:54.880 --> 00:33:07.519
So, along with this, open AI claims 68% fewer responses containing factual errors versus GPT 5.6 instant on areas such as financial, medical, and legal prompts.
00:33:07.759 --> 00:33:10.640
So Zol just dropped its new.
00:33:10.799 --> 00:33:11.839
So who has access?
00:33:11.920 --> 00:33:15.440
Well, literally one billion users, yes.
00:33:15.599 --> 00:33:19.920
So uh one billion people have better models altogether.
00:33:20.079 --> 00:33:28.319
So this is not only free plans getting unlimited GPT 5.6 Luna, which I still don't understand how that's possible, right?
00:33:28.559 --> 00:33:32.160
How is it possible that, like, okay, let's just say anthropic, right?
00:33:32.240 --> 00:33:44.000
Let's just say for whatever reason, obviously they the you know, opus five um and fable five are like you know, one A, one B or one B, one C, depending on how you look at it, with GPT 5.6 soul.
00:33:44.160 --> 00:33:49.200
But after that, right, it's it's GPT 5, uh GPT or sorry, sonnet five, right?
00:33:49.920 --> 00:34:00.079
On Sonnet 5, you on a paid plan, right, in your five-hour window, you might get 10 to 20 prompts, right?
00:34:01.039 --> 00:34:02.160
On a paid plan.
00:34:02.480 --> 00:34:07.680
So now on a free plan on Chat GPT, you get unlimited.
00:34:08.000 --> 00:34:17.360
So this again, I like I know I sound like a broken record, and you know, there's plenty of stories uh, you know, um chronicling this online.
00:34:17.519 --> 00:34:20.400
It's just been a bad couple of weeks for anthropic, right?
00:34:20.480 --> 00:34:28.239
Between all these new open models coming out, Deep Seek comes out uh essentially cheap, like free, right?
00:34:28.320 --> 00:34:30.480
Uh OpenAI, we talked about this last week.
00:34:30.559 --> 00:34:34.639
They essentially made their GVD 5.6 Luna essentially free.
00:34:34.800 --> 00:34:38.559
Uh so yeah, it hasn't been a good couple of weeks for Anthropic.
00:34:38.719 --> 00:34:40.960
So uh we'll see where their number lands.
00:34:41.119 --> 00:34:44.559
They're expected to go public maybe in like as soon as two months.
00:34:44.639 --> 00:34:46.880
So that one uh should be interesting to see.
00:34:46.960 --> 00:34:49.119
So uh who's gonna find it valuable?
00:34:49.199 --> 00:34:51.440
Why is this useful from open AI?
00:34:51.599 --> 00:34:52.960
Well, uh a couple things.
00:34:53.039 --> 00:35:00.000
Well, there's no more having to choose between instant and thinking uh for paid users, it's just one model with a dial.
00:35:00.159 --> 00:35:04.800
And then obviously the unlimited free chat absolutely bonkers, right?
00:35:04.880 --> 00:35:21.440
This is, I think, open AI uh starting to in uh inch closer to you know being able to provide this level of intelligence to you know everyone in the world, which it seems like you know, that's what their goal is, you know, bringing AGI to the to the masses, right?
00:35:21.599 --> 00:35:23.840
Um, I still cannot get over.
00:35:24.480 --> 00:35:28.239
Like when I read that, when I saw this announcement, I'm like, this can't be right.
00:35:28.480 --> 00:35:34.480
Like, I'm like un unlimited GBD 5.6 because usually free users, right?
00:35:34.639 --> 00:35:39.679
Because it's all about compute, you know, usually free users are one tier below, right?
00:35:39.760 --> 00:35:43.440
Which is how it was, you know, yesterday, right?
00:35:43.679 --> 00:35:53.199
Uh you know, free users were on GBD 5.5, and that's how it's usually been, because free users make up the majority of that billion users.
00:35:53.360 --> 00:35:59.599
So um open AI had in in terms of post-training and inference, they've cracked something, right?
00:35:59.760 --> 00:36:02.639
I had kind of my uh recursive self-improvement show.
00:36:02.719 --> 00:36:10.239
It's like, yeah, we we no one has full RSI, but it seems like open AI's got something figured out.
00:36:10.480 --> 00:36:24.000
The fact that they lowered the price of GPD 5.6 Luna by 80%, and now they are giving a uh frontier level model uh away to a billion people unlimited, unlimited for free.
00:36:24.159 --> 00:36:24.800
Doesn't make sense.
00:36:24.880 --> 00:36:30.880
All right, anyways, I'll stop losing my marbles and let you get to go using all of these new AI features.
00:36:30.960 --> 00:36:31.840
I hope this was helpful.
00:36:31.920 --> 00:36:32.880
If so, let me know.
00:36:33.039 --> 00:36:38.559
Subscribe to the podcast on Apple and Spotify, then go to your everydayai.com.
00:36:38.719 --> 00:36:39.519
Thanks for tuning in.
00:36:39.599 --> 00:36:42.719
I'll see you Monday and next week for more everyday AI.
00:36:42.800 --> 00:36:43.760
Thanks, y'all.
