Ep 785: What’s new in Gemini 3.5 Flash, Google Omni and Antigravity 2.0: Hands On With the latest from Google I/O

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Google literally had a blog post called 100 Things We Announced at I/O 2026, and most of them were obviously AI-based. And on Wednesdays, we go hands-on with the latest and greatest AI updates in our AI at Work on Wednesday series. Obviously, we're not going to dive into all 100 updates that we got from Google's I/O conference. These shows are already long enough. But of the dozen or so key AI updates that are available to most users right now, 3 seem to be getting the most attention out of the gate. For better or worse, and that's the new Gemini 3.5 Flash, the Gemini Omni Flash video model, and the new Anti-Gravity 2.0 app. It's— although the benchmarks and the keynote and the initial PR coverage might have you believing that everything that came out of Google I/O was all token-friendly sparkling unicorns, that was not the case, as there's some new wrinkles in the Google Gemini armor that weren't there previously. So we'll fill you in on all of that and give the hands-on live demo treatment to Gemini 3 Flash, Gemini Omni Flash, and Anti-Gravity 2.0 on today's show. What could go wrong? Well, let's get to it and put AI to work this Wednesday and highlight the good and the bad of these 3 Gemini standouts and show you hands-on in live how to use them. All right. So if you're new here, this is Everyday AI, but I want to zoom out and talk about the big picture. Gemini 3.5 Flash is actually a really good model, but it might not mean what you think it means if you are, well, familiar with the Google Gemini family. So Gemini 3.5 Flash is a very capable and technically really good model. It's actually Google's best AI model out there, right? But the name Flash used to mean fast and cheap, but it doesn't really mean that anymore because now it's just really fast but actually kind of expensive. And it is Google's best model. Gemini Omni is an impressive anything-to-anything world model, but I think people are mostly judging it as an AI video model and they're only looking at its ability to output video, which I think is a big miss. And then Anti-Gravity 2.0 is Google's first mainstream foray at a desktop app for agentic coding. Yet it's extremely underwhelming, and I'm trying to find any reason I can to use it, but I can't. So on today's show, stick around with me for the next 25 to 30 minutes. Here's what you're going to learn. You're going to know why Google's new best model actually got both better and a lot worse for both Gemini and API users. You're going to know why OMNI is better than the eye tests and side-by-side comparisons actually suggest. And I'm gonna tell you why you should probably just skip Anti-Gravity 2.0 for now. All right, let's get to it. Welcome to Everyday AI. If you're new here, my name's Jordan Wilson, and well, we do this every day. This is your daily unedited, unscripted livestream podcast and free daily newsletter helping everyday business leaders like you and me not just keep up with what's happening in the world of AI, but how we can take all this information and use it to actually grow our companies and our careers. So that is what our, you know, Wednesday show really does. I say, let's go. I'm going to be sharing my screen, my desktop live, and we're going to learn. This is one of those, and most Wednesdays, Wednesday shows might be a little better for our podcast audience to watch the video version. So you can always do that on our website at youreverydayai.com. However, I am going to do my best to narrate exactly what's happening on the screen, just in case you're like me and you like podcasts the old-fashioned way, just with the audio. All right. And make sure you do check out today's News, uh, newsletter. We're going to be recapping the highlights of today's show as well as all of the other AI news you need to know to be the smartest person in AI at your company. All right, so like I said, a ton new happening, uh, at Google. So, uh, they had their I/O conference now, uh, about, uh, 2 weeks ago actually, or no, 1 week ago. Time flies, time flies when you're, uh, in Codex all day. Uh, so Their conference was last week and they announced a ton. But let me just be honest, this isn't the— these 3 products are not the 3 that I'm most excited about. Probably the most that I'm excited about is Gemini Spark. It's new, Google's new kind of always-on agent. But not everything they announced is available to everyone. So again, I've kind of joked around about this, but this is the reality. There's 100 new AI updates that they announced. And number one, they're not available to everyone. It's across different tiers, depends on what plan you have, what, you know, settings you have enabled in your Workspace, if they're on Workspace, if you're using a personal Gmail, right? So there's so many, you know, hoops and fences you have to jump over and through just to understand what's going on. But I will tell you this, a lot of the more impressive or more kind of headline-grabbing updates that Google talked about aren't available to the masses, like the Gemini Spark agent, which is essentially, for lack of better terms, Google's version at an online or cloud version of OpenClaw, right? That's kind of like what everyone's been announcing, right? Here's an always-on autonomous agent that runs in the background and connects to all of your data, but it's secure, right? So that's what Gemini Spark is, but it's not available to everyone, right? There's another feature that I actually only have access to in my personal account. So it's not really worth showcasing in my work account, which is called the Daily Brief. Right. So there's so many dozens of actually really good and really exciting use cases I would personally love to show. But it's kind of hard because not everything is announced now. You know, some things are rolling out in the coming weeks, the coming months, you know, this summer. Right. So I don't know, you almost have to have an agent. And I actually did this. I had Codex go through and check the 100 blog posts and map everything out. Uh, and even there, it's still so confusing, right? Because as an example, I keep talking about Gemini Spark. It's, uh, their Spark agent because I want to use it, right? So, uh, if you have a Gemini Business plan, uh, the release notes said that it was available in Gemini Enterprise, but on the main homepage promoting everything Gemini Enterprise, it's not on there, right? So a lot of confusion on what's available on what price tier, when and how and who. And, you know, do I have to know someone at Google to get access to all of these? I do know people at Google and I still don't even understand which account I should use anyways. All right, let's get into it. Enough of my, my little riff there. So actually, we're going to start live. Do this one a little different. What could go wrong? Livestream audience, if you could let me know, All right, I'm gonna be sharing my entire screen here. Let me just make sure I'm not sharing anything too personally identifiable here, right? All right, so podcast audience, I am sharing my entire screen here and I have a couple of things popped up. We're gonna start some things live now and then we're gonna get back to some of the details on what's new, bullet point, all that good stuff. So what we're going to do is we are going to start in anti-gravity. So what is anti-gravity? Uh, well, to me, it's a piece of software looking for a problem to solve. Uh, because I'll say this, most people are not Gemini exclusive, right? A lot of people, I think, are using Gemini and something else, right? So whether that's Gemini and Copilot, Gemini and ChatGPT, uh, Gemini and Claude, right? If you are using gemini.google.com, right? So the Gemini chatbot, if you have that $20 Pro plan or $100, you know, the new $100 Ultra plan, Google also brought down the $200 Ultra plan down from $250, right? So if you have any of those plans, I'm guessing you probably use something else. So if you have a $20 a month ChatGPT plan, then you have access to Codex. If you have the $20 Claude AI plan, you have access to Claude Code and Claude Cowork. So this is— Anti-Gravity is kind of Google's version, for lack of better terms, of something like Codex or like Claude Code. So it is their kind of fully integrated agentic desktop system where you can run agents in parallel and kind of connect to your data. But I'm not even really sure how. Anyways, so on my screen I'm showing this the usage because this is one of the big downsides in some of my earlier testing. The usage is actually fairly bad now, right? It's not Anthropic bad, but we're gonna be following along live. So I had to switch over to one of my other paid accounts. So I'm on a $20 a month account. You'll see on my screen here inside of Antigravity, I am fresh, I have 100% remaining. And then if we look into my inside Google Gemini, gemini.google.com, same thing, I have 100% of my usage remaining. So I'm doing this live. We'll see because of my testing, it was really, really bad. All right. So first let me give everyone a quick, uh, kind of tour around Antigravity. It's going to be quick because at least right now there's not a lot to see, right? If you are a heavy Codex user like me or Claude Code, Claude Cowork, Cursor, et cetera. If you're using any desktop agent tool, you're going to look at Antigravity and be like, well, where is everything? Well, this is it. So it is very beta-ish. There's not a lot to see, but I will tell you one big advantage that you can, well, with Google's Anti-Gravity 2.0 desktop app versus something like Codex or Cloud Code is, well, you actually have different models that you can use. All right, so it is all based on your plan. So aside from the new Gemini 3.5 Flash, which is the models we're going to be using, obviously you have the other Google Gemini model, Google Gemini 3.1 Pro, but you also have Claude Sonnet 4.6, Claude Opus 4.6. You have the open source version of GPT-OS-S, the 120B. So this is actually OpenAI's open source model. So I do like that with Google, you have, you can use different models from different providers, right? Obviously you can do that with a lot of third party, you know, or Cursor, you know, but within Google, it's nice that you're not just quote unquote forced to use Gemini models, although obviously the Gemini models are really good. All right. So just kind of starting with the unique kind of call out there. Aside from that, I don't know. To me, there's nothing really unique, at least right now inside Antigravity. If you look at the interface, you know, people are joking around. I already joked around about this and Google's already been kind of dragged through this. It looks very Codex-esque, right? Even in their launch video, the 2 minutes, it was like a 16-minute long video, uh, going over anti-gravity. Literally, they had a desktop icon with the word Codex in it. So maybe they drew a little bit too much inspiration from Codex because it literally looks just like Codex. Anyways, on the left-hand side, you can start new conversations, uh, you can create new projects, uh, new chats that are not part of projects. So again, literally just like Codex and, uh, kind of the big difference here is you can give Google Antigravity read/write access to any folder or folders on your computer, right? That's one big kind of advantage of these kind of desktop coding agents, as well as you have scheduled tasks. So that's the other thing. So you can, you know, let's just say you wanted to clean your downloads folder. I know this is a lame example. Let's say you wanted to clean your downloads folder once a week on Sunday nights. You wake up you know, Monday morning refreshed, right? So you can schedule something like that and it will run all of these things on a schedule. All right. So if I go into a new chat here. All right. So you can create a new project. All right. And this is going to pull up all your folders right there. So I'm not going to do that right now. I'm just going to go into a folder I currently have set up and I'm going to start a new conversation. So when you start a new conversation for this one, it has access to the download folder, and then I can click backslash for different actions. So it does have a goal mode, which is nice. You can run things on a schedule, browser. So, and then there's also built-in skills. So by using the backslash command, you get some pre-scheduled kind of actions as well as skills that you can use. All right. Or if you click the button, you can reference other conversations or different agent rules as an example. All right. So nothing there to really see. Very straightforward. Um, the thing that I didn't fully understand, and I'm like, wait, I don't understand this, right? I'm clicking around the account, the permissions, the appearance, the models, the customization, the browser, all of these side panel settings. And I'm like, okay, where do I kind of connect my, my Gmail, my Google Drive, my Google Sheets, my Google Calendar, all these Google products that you would just assume? And I'm like, they're not here. I don't know, I keep thinking like, oh, I have to be wrong on that. But, you know, I'm looking, I'm searching, I don't see them, which makes to me absolutely no sense, right? Obviously, you know, Codex, Claude Code, Claude Cowork, they all have connectors, cursor, right? Apps, connectors, whatever you want to call it, integrations. I don't know. At least I'm looking at Google Antigravity, nothing. Anyways, let's put it to the test. All right. So here's what we're going to do. I have a prompt ready to go and we're going to check back on this in a little bit and we're going to check at the usage. So all I said here, just so you can kind of see what's cooking and kind of see Antigravity starting to work here. I said carefully look up the Start Here series by Everyday AI and carefully and meticulously denote the main characteristics, takeaway, etc. of each episode. Then create a simple but executable web app that is interactive, filterable, and sortable for a novice user user who is wanting an easier way to navigate through the series. Make sure to include a way that ultimately drives people to Everyday AI's official resources for the Start Here series. All right, then I'm saying at the end of your output, please give me the exact way to run or launch this. All right, you'll see, uh, anti-gravity, it can't be done yet. All right, it might already be done, uh, but let's see if it actually properly Nope. Okay. It's still going. It's just breaking this down into separate tasks. So I would love to be able to run another test in parallel, but in my testing beforehand, when I did that, actually none of it worked because it all timed out. It all, it hit me over my usage limit. So I will make sure to do just that one. And then we'll go back in later and see if we can't do another anti-gravity test. All right, so stepping away from the live demo for just a second. All right, well, I was going to say let's go back and check on it, but it looks like it's, looks like it's already done. That wasn't, that wasn't too bad. I thought it was going to work for a little bit longer. All right, so I'm just making sure I can open it up here and there we go. Alright, so AI moves too fast to follow, but you're expected to keep up. Otherwise, your career or company might lag behind while AI-native competitors leap ahead. But you don't have 10 hours a day to understand it all. That's what I do for you. But after 700+ episodes of Everyday AI, the most common questions I get is, Where do I start? That's why we created the Start Here series, an ongoing podcast series of more than a dozen episodes you can listen to in order. It covers the AI basics for beginners and sharpens the skills of AI champions pushing their companies forward. In the ongoing series, we explain complex trends in simple language that you can turn into action. There's 3 ways to jump in. Number 1, go scroll back to the first one in episode 691. Number 2, tap the link in your show notes at any time for the Start Here series, or you can just go to starthereseries.com, which also gives you free access to our inner circle community where you can connect with other business leaders doing the same. The Start Here series will slow down the pace of AI so you can get ahead. Let's take a look here. So it created a nice, you know, kind of localhost, right? So this isn't a hosted website or anything like that, but it does have everything here, right? So it's a nice kind of interactive webpage. It went through, I'm kind of looking here to make sure that it got all of them, all right? So unfortunately, when I'm filtering, I— okay, here we go. All categories. So first I'm seeing we have our Start Here series, which is the podcast for beginners and people who are trying to grow in their knowledge of AI. And it looks so far pretty good. Everything in here, it looks like it got all 25 episodes. I'm looking at the title, right? So Volume 1. Was Generative AI: How It Works and Why It Matters in 2026 More Than Ever. That is correct, right? And then let's see the latest build by Partner Await, the 4-Layer Stack Decision Framework for 2026. So this is good. So it not only got all of the volumes correct, the titles correct, pretty good overall. The design leaves a lot to be desired. I didn't give it any, you know, certain prompting or anything like that. So it looks very AI, right? Like very AI generated. I wouldn't call this, you know, visual slop, but if you've used any, you know, anything on the Vibe Coding side to create anything, it looks very Vibe Code sloppy, right? So the dark mode with the, you know, the neon gradients. So not terrible, but, you know, nothing I would actually want to use or to put out there. And then when you click on something, so let's go click on our read us— sorry, our latest episode, the Build, Buy, Partner, or Wait. So I click in this, there's a nice little modal. It has some additional key takeaways and action points, some related tags. It has the episode number, the Gitfo full playlist, and then the listen or show notes. So let's actually see. So the Gitfo playlist does not actually work. That is not going to get us the full playlist. And then the listen show notes, let's see where that takes us. Okay, so it just took us— I didn't even know this page existed on our website. So those buttons, in theory, it took you somewhere, but it did not actually take you to where you wanted to go. So, you know, as an example, if I was using Codex, it would have, number one, taken a lot longer. And I know because I've done this test before, this is one thing I'm, you know, thinking about making part of my, you know, testing series that I do a lot because it actually requires a lot of different tool calling, right? It has to go to the website a lot of times. It has to read at least 25 different pages, right? But I know Codex actually, when it builds these things, it clicks through and it can test the links. I think Claude Code does a pretty good job in that as well. So I wouldn't say the, the AntiGravity version failed here by any means. It looks fine, but some, some things that don't really work. But the fact that I'm critiquing something that looks like this, right? I have a little bit of a background in, you know, web design, but pre-AI to build something like this, it would have taken, I don't know, would have taken at least 30 hours of nonstop work, right? Even if I'm working with a template to be able to go through, grab all this information, put it up on there, you know, it is nice, smooth. There's animations, there's filters. So it is almost like a small database, right? It is 25 episodes with correct-ish links out, uh, you know, additional information, you know, the right episode number, related tags. So overall pretty good. All right, let's get back though to going over some of the details of Gemini 3.5 Flash, um, the Omni Flash video model, and then also Antigravity. And then we'll go back hands-on with Gemini Flash 3.5 and the Omni video or any-to-any model. So real quick, Gemini 3.5 Flash, this is not going to be Google's final 3.5 model. So Google did announce that they next month will be coming out with a Gemini 3.5 Pro model or Gemini Pro 3.5. So we will see what happens with 3.5. Flash, but right now it is actually better. The Flash version is better than Gemini 3.1 Pro, which we haven't seen that yet, right? So always Google Gemini 3, you know, Gemini X Pro is always Google's best model. So this is a little weird because the Flash kind of moniker in the Flash series up until now has always been about being super fast and super cheap to use. It is even faster. Than before, but it is not cheap. All right. So depending on what you're looking at, you know, you might actually be paying 10 to 20x more if you are using this on the API side than previous, you know, Flash versions. And I do have some examples of that, but it does outperform Gemini 3.1 Pro right now on a lot of different benchmarks. It's available now, whether you're using the Gemini app It is also available in search. So the new AI mode is powered by Gemini 3.5 Flash, which is really cool. And also it's available on the API as well. So speaking of those benchmarks, yeah, it's pretty good here. So showing a, a screenshot here of all the different benchmarks from Google's announcement blog post and you'll see for the most part, it beats Gemini 3.1 Pro in every single benchmark category. Some of them are actually pretty notable, right? Like if you look at TerminalBench, you know, which is important for agentic coding, you know, pretty big jump up from Gemini 3.1 Pro. The same thing with some of these more agentic, you know, the agentic benchmarks. It does really, really well. I mean, we'll see here. I think Really what Google is, I think, going to be using Gemini 3.5 Flash for in the future, obviously after Gemini 3.5 Pro, is, well, to run long-running agents. What's interesting here in this Google benchmark sheet that they released against the Gemini, so Gemini 3.5 Flash, Gemini 3 Flash, Gemini 3 Pro, Claude Sonnet 4.7, Claude Opus 4— sorry, Claude Sonnet 4.6, Claude Opus 4.7, and GPT-5.5. The interesting thing is Gemini 3 Flash, uh, even though these are cherry-picked benchmarks, obviously, because there's dozens of benchmarks out there, so they cherry-pick the ones that, you know, they thought told a good story. But interestingly enough, GPT-5.5 and Gemini 3.5 Flash are winning the same amount of benchmarks on the ones they chose. So, you know, when people are like, like, Jordan, like, you're crazy. You're talking about Codex and GPT-5.5 too much. I'm like, no, it's right now. It's just the best and it's not really close. And I continue to say this. I don't think Gemini 3.5 Flash overall is anywhere in the realm of GPT-5.5. And when we look at the artificial analysis, Artificial Analysis Intelligence Index, you'll see that holds true, right? So Artificial Analysis, a great kind of benchmark conglomerator, right? It just throws all these different benchmarks together and gives all the models an intelligence score. So Gemini 3.5 Flash, you know, did fairly well considering it is a flash model. So it is still on this benchmark at least well behind Gemini 3.1 Pro. and still very, very far behind the best overall model in the world, which is still, uh, GPT-5.5 Extra High or GPT-5.5 Pro. So that is still very far ahead of, you know, Claude Opus 4.7 or Gemini 3.1 Pro. But the thing that you have to call out is Gemini 3.5 Flash is no longer cheap. It is actually very expensive compared to anything in the Flash family. I actually think it would have been— it's a more apt name or a more accurate name. It just— if they called this like 3.2 Pro or 3.1 Pro Fast, because the only difference, right, it is not anything like the Flash models from before except that it's fast, but it's actually really expensive. So the artificial analysis has the intelligence versus cost kind of index, right? So I have this kind of screenshot here that I'm showing to our audience. And for the most part, you want to have a model that is intelligent. So on one axis, it has the artificial analysis intelligence index score, and then on the other one, it's the cost to run the index. So this is a set of tests, and then Artificial Analysis literally runs all of these tests via the API and adds up how much it costs to essentially run all these tests to get these scores, right? So essentially, you want to be in the upper left-hand quadrant, which is a very smart model, but doesn't cost a lot of money to run. So obviously, you want to be in the upper left-hand quadrant. So Gemini 3.1 Pro, well, it was, right? It's in the upper left-hand quadrant along with models like GPT-5.5 Medium. You have some open source models that are kind of on the edge there. But for the most part, it was Gemini 3.1 Pro, GPT-5.5 Medium, QN.37 Max. Um, and then there's some on the edge there, but Gemini 3.5 Flash, again, it's not up there in terms of intelligence. It's not a, you know, top 5, uh, you know, intelligence model out there right now overall according to artificial analysis, but it's actually fairly expensive. It was much more expensive to run these tests on Gemini 3.5 Flash than it was, uh, on Gemini 3.1 Pro. That's because it is token hungry. Gemini 3.5 Flash eats way more tokens than it needs to. And why is that important? Well, it's important both on the API side. So if you are building on these models in your company, if you're using these models, which is gonna become easier and easier and probably more relevant and more prominent in the future, because I think kind of this whole, you know, token subsidy thing is going away, right? So Google has kind of gotten rid of it. Like I talked about, they've introduced these new usage limits, which they didn't really have before. Anthropic limits aren't really usable if you're paying for a, you know, $20 or $100 or $200 a month subscription. So the only out of the big names, right, is just OpenAI that still has these, you know, usage limits that seem like they're, oh my gosh, it seems like they're unlimited. So anyways, intelligence versus cost to run Gemini 3.5 Flash costs a lot of money if you're using it on the API side in terms of the intelligence that it can output. Per dollar. All right. So let's talk about Google's new anything-to-anything model, which for the most part, even myself, right, we're calling it their new video model because it does replace their Veo series. But the Google Omni Flash is Google's new multimodal model that combines Gemini's reasoning, reasoning with generative video creation. So what makes this different than your run-of-the-mill you know, text-in, video-out model, well, is it accepts a lot of different formats on the front end. It accepts image, audio, video, text inputs, as well as outputting grounded video that you can actually edit. So this does replace Veo, and it's available in the Gemini app and Google Flow for AI Plus, Pro, and Ultra subscribers. You can only generate 10-second clips at a time, but there is multi-turn conversational editing at launch, which is really cool. some things that really stood out with Omni versus the VO3.1, the last video model from Google, is improved character consistency, preserving identity and voice access across every scene, and it excels at video editing like replacing objects, backgrounds, foregrounds, etc. So, um, this is the thing. This is a world model first, right? One thing that I think Google has always been on the cutting edge on, and I don't know if any of the major labs are going to be able to catch them anytime soon is just being multimodal by default, right? This is something I told you guys like 2 years ago. I'm like, why aren't more people talking about like, you know, using the— I think it was Gemini 2.5. Gemini 2.5 Pro might have been the first model that could ingest video, right? Ingesting video in Google's AI Studio, not just being able to, you know, actually have that as an upload source, but it could look at it every single frame and tell you what's going on, right? So Google has always had this huge benefit that I think is going to slow slowly reap the rewards in the years to come when we talk about world models, robotics, embodied AI, etc., right? But this is where the Omni Flash version might eventually show some of the fruits of those labor, right? As well as there will be other versions of Gemini Omni according to Google. So we might get either a Gemini Omni, you know, versus just the Flash version, or, you know, a Gemini Omni Pro. So, and then let's quickly bullet point what's new in the, uh, Anti-Gravity 2.0. So this is Google's standalone agent-first desktop app for Mac, Linux, and Windows with no IDE required. It is the central home for orchestrating multiple AI agents and executing tasks in parallel. We'll see about that because my test did not allow me to on a paid plan. Uh, it can dynamically add sub-agents. You can schedule tasks in native. Voice commands. It's a suite that also includes the new Antigravity CLI and SDK and managed agents. And it is powered by Gemini 3.5 Flash and co-developed using Antigravity itself. And apparently maybe Codex as inspiration. And one of the bummers, one of the downsides, a lot of people are upset about this. I'm not really a big command line interface CLI user, but it does replace the Gemini CLI, which is going to sunset for consumer users next month. So June 18th. And a lot of people are upset about that because Gemini CLI was a, I think, a great product if you were more of a technical command line user and it was open source. So Google shutting down a popular open source project and instead making it a closed source. So, all right, that's enough for the bullet points. Let's go back live, shall we? All right. So I'm going to go ahead and bring back up antigravity. All right. Livestream audience, let me know if we got it back. Do we got it back? Here we go. All right. So we got our one. All right. And I'm going to get another antigravity prompt ready. We're not going to sit around and watch it, but I'm curious what happened with our— what happened with our limits here. So let's go to our models. Okay, so this is very strange. All right. All our models still say 100% remaining. So maybe I was running into a bug because when I did this one prompt, it exhausted all of my limits on a paid plan before. But you'll see now. Okay, good. So I'm clicking refresh again. All right, so maybe we're good. I'm actually going to literally close this down and reopen AntiGravity just to double check. All right. All right. So apparently we still have 100% usage remaining. So I'm actually going to just go ahead and do one more, one more little prompt there. We'll check in on it at the end, but let's go ahead and do the same now for Google Gemini. So I always have kind of a funky little rubric that I take some models through just to really see, are they smart? Are they fast? How do they work? All that good stuff. All right. So one other thing about Google Gemini, the new app, it actually looks really nice. I love the new user interface. They did redesign it from the ground up. Which I really like. So some things if you are new to Google Gemini, they do have this new daily brief. I'm not going to be able to bring that up because this is my personal Gmail account because that's the only one where it's available. And my other workspace account ran out of limits from doing like nothing. So on the left-hand side, this is where you can start a new chat. You can have the new daily brief if you do have that. Images with Veo or sorry, images with Nano Banana, videos now with the Omni model, your library, your Google Gems, which are similar to GPTs, right? Just smaller versions of Google Gemini that you can create. You have your notebook with your Notebook LLM integration, and then your recent chats. All right. So in the main interface, I'm just going to go ahead. I have a series of prompts that I like to run. One thing. All right. I'm going to refresh this, see if this works. Okay. Interesting. All right. One thing that's confusing to me. Is even the model selector is not consistent across Gemini interfaces. So with my personal Gmail Gemini account, which I have up, right, my options are 3, right, 3, 5, Flash, and then with thinking levels. Okay, cool. Right. And then I go to my paid version of Gemini on my workspace plan, and I don't have that. All I have now is 3.5 Flash and 3.5 Thinking. So that's one thing I think the Google team needs to tackle, right? Especially for people who are using, right, like myself. And I know there's millions of other people, I'm sure, that are using Google Gemini on their personal Gmail account, in a work account. If you want to be able to replicate things, test things, run things through use cases as a front-end user, which is really important as teams start moving to more front-end surfaces, you have to be able to let people understand, number one, what model they're using. Is there a knob or a control for reasoning effort? So at least me, I've been thinking, I've been reading about it. It's not marked anywhere. It's confusing. All right. Anyways, so for this, I'm going to be using the 3.5 Flash with, you know what, I'm going to— all right, I'm just going to do— what am I going to do? Let's just go ahead and do extended. All right, we'll see if this works. And right now, uh, my usage is 0%, so I have 0% used. All right, so my fun little rubric, it's about eating bananas and stuff. All right, so I want to see, uh, the correct answer for this is 5 apples and 3 bananas. So I just said, I just woke up today with 6 apples and 3 bananas. Yesterday I ate a banana and 2 apples. This morning I will eat 1 apple and no bananas. However, I don't really like apples and 1 banana may turn brown tomorrow. Assuming nothing else changes, how many apples and bananas will I have tonight. All right, so the good thing is, well, Google Gemini got this correct, so that's good to know. All right, I did want to show you this because one new feature that Google has that they didn't really talk about, which I like-ish, I wish they went a little bit further, is you can finally see a little bit of the Gemini chain of thought. Not a lot. What is chain of thought? Well, that shows you how a model tackles a problem and what tools it calls, right? So is it uploading? Is it accessing certain emails? Right. If I was connecting to my emails here, that would make sense. Or what emails is it looking at? What documents is it looking at? What websites is it going to? Is it using Python to write this code or is it doing the math in its head? Right. So as an example, now I can click these three little dots at the end of the prompt answer and it says more. And then I can see, see thinking steps. So unfortunately, it's, it's a bad chain of thought. You don't really get to see anything. But before, Google Gemini had absolutely nothing. So sometimes when I really push and add a lot of files and give it a complex prompt, I can see a little bit, at least about some of the websites it goes to and some of the thinking steps. But it is nowhere near what you get with ChatGPT and with Claude. I actually think right now Claude has a slight advantage in, in having a more transparent chain of thought that's easy to read. Uh, ChatGPT's is ultimately better, but it's a little more difficult from a user experience to get to it. So, yeah, so, uh, Google, they kind of went for it, but it kind of didn't work as well. All right. A couple little, we'll do these fun little things. All right. This is one that most, uh, large language models get confused at. A man and his dog are standing on one side of the river. There's a boat with enough room for one human and one animal. How can a man get across with his dog in the fewest number of trips? All right, Google Gemini 3.5 Flash got it right. One trip, right? I don't know why, or maybe these things are finally in the training data because I don't know. I talk about them a lot. I'm sure other people give similar prompts. All right, next one. If it takes 3 hours to dry 10 t-shirts in the sun, how long will it take to dry 30 t-shirts in the sun? Takes 3 hours. Got it correct. All right. So, so far, some of these trick little prompts, Gemini 3 Flash doing a good job. If you have a single match and you walk into a room with an oil lamp, candle, and a fireplace, which do you light first? The match. That is correct. All right. Next one. What color is an airplane's Black box. All right, the answer here is orange. Let's see how it does. All right, got it right, bright orange. I'm going to jump over to usage, FYI, click my refresh. I swear, I swear, y'all, it only says 1%, which is great, which is great. Uh, before when I was doing these tests, it was It's really not letting me do hardly anything on my other account. All right. Anyways, let's keep going. And then I'm going to say, please give me 7 jokes that end in the word blue. 2 should be about animals. 3 should be about some other topic in the body of this chat. And you can make up the other 2. All right, let's go ahead. We'll give it a second. I'm going to get my— all right, let's see if it got this right. Animal joke. What do you get when a sad puppy falls into a bucket of paint? A golden retriever that is completely blue. So no model has ever been able to make these actually funny. But I will say that this is coherent. It's ending in blue. I'm looking at the rest. It's taking some from the fruit topic. Both end in blue. Correct. Okay. And then the made-up jokes all end in blue as well. We'll see if the made-up ones are actually funny. Why did the jazz musician stare at the ocean all day? He wanted to write a song that felt blue. All right. So no jokes are actually funny, but they're at least coherent and they follow the rules. Last but not least, in my silly little rubric, a box is locked with a 3-digit numerical code. All we know is that all digits are different. The sum of all digits is 9 and the digit in the middle is the highest. What is the code? This is a trick question. Because there's actually multiple possibilities. So let's see here. There we go. Google Gemini 3.5 Flash says there is actually more than one correct answer based on the clues provided. There are 14 possible codes. All right, let's see. Codes that start with a non-zero digit. All right, they all add up to 9. Middle number the highest. Yeah, so 180, that works. 270, that works. 360. 4, 5, 0. Yeah, so it did a good job. It got through that. So now just for fun, I'm going to go ahead and open Canvas mode and I'm going to say build all of my prompts and results into an interactive website using Canvas mode. All right. So obviously all of the old and previous features of Google Gemini, obviously still work with the new model. I haven't used Canvas mode as much because I've noticed on my work plan, which is the one that I primarily use that's connected to all of my work data, my limits were disappearing like peanut M&Ms, like upstairs, right? They were just vanishing. All right. Anyways, we're going to give that a second. We'll check back in on that interactive on that interactive app there. And then we'll just do our last kind of test here, which is looking at the Gemini Omni. All right. So, so let's go ahead. I'm going to click the create video and let me make sure. All I'm saying is explain a large language model, but using basketball terms. All right. So I do believe right now it is an 8-second limit. I was trying a video to video, a video to video example. For whatever reason, this has been going for like 2 hours. So I don't think this one is going to work. I don't think that our— okay, interesting. So I said, explain a large language model, but using basketball terms. And it said, I can only generate videos. Uh, try another prompt. All right, so I'm gonna redo this even though— all right, I'm gonna say, create a video that explains, all right, how a large language model work but using basketball terms. We'll do that and see if that works where the first one didn't. All right, so let's jump in. All right, there we go, we have our interactive website version here. Done. All right, so this is kind of cool. So it did— it created a nice little interactive website for the prompts that I gave it with the correct answer as well as an explanation. Right. Which is pretty cool. Right. So the river crossing, you know, it says on the left bank, there's a man and his dog with two emojis of each. The boat. It clearly shows that there's two seats in the boat, and then it kind of shows that you can, um, oh, this is, wait, this is actually pretty cool. I can click on them. All right, that's kind of cool. Uh, load the emojis into the boat and then click the sail across. All right, and then it goes the other side. So, you know, I, I've, I've never been, uh, shy about how much I love Google Gemini Canvas Mode. It is one of my favorite and most used kind of little AI modes across. I love the Gemini mode. It's great with front end. It's fast and it usually works way better than Artifacts or even the canvas mode inside of ChatGPT. All right. So pretty cool there. Let's check on our video generation. All right. Here we go. We have— let me see. I actually have to— so we can get some audio here. I have to stop this screen share and then I have to share the screen only using tabs. So let's go ahead and do that. So maybe we can all hear. Okay, I mean, oh, okay, I can share system audio. No, I can't. All right, this is, I had to do this in my other plan. All right, so I'm gonna have to download this video and then I'm gonna have to open it in a different window. This is why I love doing this stuff live Y'all, I love doing this stuff live. All right, let's try this again. All right, and then let's drop in the video I just downloaded explaining. All right, let's try this here. Let's go Chrome tab. Thanks for your patience, y'all. Here we go, Chrome tab. All right, we'll see how a large language model works in basketball terms from a visual perspective. We have a basketball player, clearly, you know, it's supposed to be the Bulls, but no Bulls. So, you know, Chicago Bulls colors, number 23, someone in an arena. I don't know if it's actually the United Center because funny enough, there's kind of Canadian banners in the background, but the physics, everything looks good. Court looks good. So let's go ahead and listen to the 10-second, can they explain large language models in 10 seconds with basketball references? Imagine a massive scouting report. An LLM reads every play ever made. It practices billions of shots to predict the next bucket. It is just ultimate teamwork. All right. So, nothing terribly inaccurate there. All right. Some of the physics were fine, right? He kind of shot a basketball, but, you know, at least you can kind of see the ball kind of goes through the rim there. And at least where he's shooting it, it's kind of a close-up. Unless he's, you know, Victor Wambiyana or whatever his name is, his shooting arc is not going to be that close. to the rim unless he's standing on his teammate's shoulder. But I mean, overall decent 10-second explanation using basketball terms, right? All right. So this one went a little long. Apologies about that. But here's the big takeaway. I think there's pros and cons here, uh, to Gemini 3.5 Flash, to Gemini Omni Flash, and to Anti-Gravity 2.0. One of the underlying downsides is, well, the new usage limits. Although in my demo today, they didn't really creep up. Thankfully, that would have been bad if I, you know, started something in Antigravity and then I couldn't have done any other demos, even in Google Gemini. But a lot of people like myself have been complaining that even on the $20 a month plan, it's again, not as bad as Anthropic Claude's limits. But I mean, I think Google Gemini used to be a buffet, right? It was an all-you-can-eat from the AI compute factory and not so much anymore. So that's a downside. The other big downside with Gemini 3.5 Flash, if you are using it on the API side to build, it's much more expensive, right? It is one of the most token inefficient models there is. So although you can look at it one way and say, oh wow, it's, it's very smart for a flash model and it's very, very fast. If you're using it, if you're just trying to swap out, uh, you know, maybe you're using Gemini Flash, uh, Gemini 3 Flash or Gemini 3.1 Flashlight, and you're saying, oh, we can just swap it out for Gemini 3.5 Flash. It's going to cost you a lot, even though it is very fast and a very smart model. It doesn't really, you know, keep to the original name. So obviously Google's, you know, what they're going to market Flash as is going to change, you know, as they see user demand change. But again, very capable model. It's good. It's fast, but more expensive for the most part. And we'll see what happens with Gemini 3.5 Pro. I would expect a much better, much more capable model, especially since the fanfare and reaction to A lot of what we saw from Google I/O has been mediocre at best. Google Omni, I think people are misjudging it. Yeah, you saw my one little example there. My video to video example didn't work for this, but I think ultimately people are only looking at and comparing it to things like Seed Dance and they're saying, hey, text input, video output. That's not what Omni is. I think this really unlocks the future of creativity. Once Google is able to serve this up to more people, maybe extend, you know, the time, I think this really changes the future of storytelling. Telling, being able to, you know, shoot a quick video and then change anything about it instantly in a consumer-friendly interface is a big step change in creativity and what's possible. Last but not least, at least on the 3 things we went over today, AntiGravity 2.0, not that good. Not going to lie, right? Yes, it can create nice front ends. It is fast-ish. But when you're comparing it to what else is out there, I'm not going to use it, right? Even I think even if I had unlimited usage of Gemini Anti-Gravity 2.0 desktop app, I'm not going to use it because it lacks what you get in Claude Code, Claude Cowork, and Codex by far. Hopefully the team is going to continue to invest in it. It does look like Anti-Gravity is a bigger priority than I would have thought, at least when you look at the end product and you're like, okay, this actually got a lot of shine and a lot of play at the I/O event, and then you look at what was initially released, there's been a lot of updates even in the past one week since it was initially released. But when you look at what was released, you're like, ah, they maybe should have slid this one out as a footnote or maybe waited to release something like this. I don't think it's there yet. It's not in the first or second tier in terms of Claude Code uh, Codex, Cursor, you know, there's a handful of other third-party ones that I think are really good and very far ahead of what we have in Antigravity. So I hope this was helpful. A longer version of putting AI to work at Wednesdays. That's what we get for trying to tackle 3 things live and run into a couple hiccups along the way. I hope this was helpful. If so, make sure you do check out the video version at youreverydayai.com. While you're there, make sure to sign up for our free daily newsletter. Uh, and thank you for tuning in. Hope to see you back tomorrow and every day for more Everyday AI. Thanks, y'all. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.