AI News: Insane Week - Gemini 3 Was Just The Start

Hey, how's it going? What have you been up to?

Did you miss me? This week's been an absolutely insane week in AI news. We had Microsoft Ignite out in San Francisco, Meta launched SAM3 and SAM3D. Xai launched Grok 4.1, OpenAI released a crazy good new coding model and so much more. But first we've got to talk about how Google launched Gemini 3 and Nano Banana 2 Pro Extra series sequel the Reckoning Tokyo Drift. So without further ado, let's jump right in.

Starting with the new Gemini 3 launch. Now I'm not going to go super deep on this one because I did make a full breakdown video on this one called Gemini 3. Rumors are confirmed. It's very good and that breaks down all the details and does a ton of demos and testing with it. However, I'm not going to leave you hanging in this video. So here's the TLDR Gemini 3 is Google's new flagship thinking model and it comes with major jumps in reasoning, coding, multimodal understanding and long context performance. Meaning you can give it a ton of text input and also get a ton of text output from it. And it basically crushed all the benchmarks to pretty much be the top dog across the board. And the days of Google actually announcing products but not shipping are seemingly over because this is available everywhere as of this week. Right now you can use this inside the Gemini web app. You can switch to Gemini 3 right now it's also already powering AI mode in Google search for Google AI Pro and Ultra subscribers in the US. Gemini 3 is live in Google's AI studio where you can actually test it for free. As far as I can tell, it's also available in the Gemini API and the Gemini cli so you can build apps and agents with it as well. On top of the model itself, Google also launched the Gemini Agent, which is a new experimental mode in the Gemini web app. It's not just chatting, it can actually take action. So like browse the web, check your Gmail, analyze your calendar, pull documents from drive, build slide decks and plan multi step tasks all on its own. And finally Google announced Anti Gravity, a brand new AI powered cross platform IDE for Mac, Windows and Linux. Now if you're not a coder and you don't know what that means, don't worry about it. Probably not super relevant to you, but this new IDE integrates directly with Gemini 3 and the Gemini Cli and is designed to help you build software faster with AI driven coding, refactoring, debugging and agentic workflows. Bottom line, Gemini 3 is here. It's smarter, it's more capable across every modality and starting today it's baked directly into search. The Gemini app, Agent Mode, AI Studio, the API, the CLI and the new Anti Gravity ide. That was a lot of acrony and here are some of the coolest things I've seen people doing with it. Dane JW here made this visualization of a neural network training right in front of him and then even got it to represent itself in 3D space. That's super cool. Vortex was able to make a full 3D RTS game. Pietro here was able to give it one image and it converted that image into a pixel perfect website from the image. So like everything that was in the image ended up on the website perfectly. Zara here made an app that, that asks her questions and then she can just sit there and talk to her computer answering these questions and then it exports a video for at the end. It's like a sort of automated video journaling app. People have been creating some absolutely wild stuff with Gemini 3, so make sure you check out my other video breakdown if you want to see even more examples as well as some examples of some of the best prompts that I've used with it. And then we've got Nano Banana Pro which came out this week from Google as well and just blew everybody's minds. This is another one that I did do a deep dive breakdown on all of the the details of it. The video is called Nano Banana Pro is here new features unlocked, so check that one out if you want all of the details plus a ton of demos and example prompts. But again, here's the TLDR if you happen to miss it. Nano Banana Pro is Google DeepMind's new state of the art image generation and editing model. Built on top of Gemini 3 Pro. It delivers studio quality designs with improved control, accurate text rendering in multiple languages and and enhanced real world knowledge. So for me the standout features were the text rendering. It's actually really, really good at rendering text on images. Like even small text and background text is readable and not sort of gibberish like we're used to. It also uses live information from Gemini3 so it's actually going to generate infographics and things like that and the text that it overlays is text that it researched for you. So you can ask it to create an infographic on a specific topic. It will research that topic and and then use the information that it found in that infographic, like really cool. Wow. Number three, you can blend up to 14 images. In my original video I mentioned you can blend up to six, but that turned out to be just the sort of pre launch information. The actual launch version allows 14 images. However they do still recommend using 5 or 6 images for the best results. You can also change the aspect ratio of any image you can to any other aspect ratio without it distorting the image. Now this seems like such a small thing, but it's super useful to take like a 16 by 9 image and then reconfigure it for a 9 by 16 image. For a different social platform, old models would have like compressed or distorted the image. Nano Banana does it perfectly. And number five, there's even more advanced camera controls so you can do things like tweak the lighting and camera angles and match the color grading from other images. And it's got style transfer so you can give it like one image that you like the style of and then another image that you want to restyle and it will take the second image and apply the style of the first image to the second image. Nanobanana Pro is going to be people's new go to platform for things like infographics, flyers, banners and converting images from one size to another for sharing across multiple social platforms. It can also output in both 2K and 4K resolution and it's available right now in the Gemini app if you turn on Thinking mode and the Nano Banana mode. Once both those are on, you're using nanobanana Pro. Free users get a limited quota of generations before getting bumped down to the OG Nano Banana. And I don't know how many generations you get. They didn't tell me. If I had to guess. It's probably a sliding scale based on demand, but I don't know for sure. It's also being made available in the Gemini API, Google AI Studio, Vertex AI and the new Anti Gravity IDE for coders. So coders and developers, they're getting it built into all the Google stuff as well. Third party platforms like Adobe, Figma and Leonardo are also getting it baked in. Now if you are going to use it in the API, it is a little bit pricey. It costs about 13 and a half cents per image or 24 cents per image if you want to generate in 4K. Compare that to less than 4 cents per image on the original Nano Banana. So this quality bump does also come with a cost bump if you're using it in the API. But here's some other cool examples of what I've seen others do with nanobanana Pro. Here's some cool images from Matvid Pro. He made like a Calvin and Hobbes comic book, and it's all legible and looks like a legit comic you'd see. Here's a temporal displacement device that he designed. A pixel art. Better call Saul, I think Google CEO posted this, like, deconstructed hamburger. And keep in mind all of the text you're seeing on this is based on research that Gemini did and then implemented into the image. Here's some examples of some really cool annotated diagrams from Sahil Apollo 11 lunar landing. Look, it sort of drew all over the image with details that, again, it researched. Here's another one of this rocket ship here, one from a McLaren MCL60F1 car. I mean, if you need any infographics, you got the tool. Here's some of my favorite generations from my other video. I took a watercolor painting and a photo of me and some of my buddies here and took the style from the watercolor and applied it to the image. I took this image of this group here and had it turn it into like an 80s synth pop band poster. Here's an image where I sketched a giant person in the background and told the model what to do with it. And here was the resulting image. I made this menu with both English and Japanese text. And what's so impressive is even this small text here is legible and not gibberish. So Nano Banana, again, super, super impressive. I did a full breakdown video on it. You definitely need to check that out. I love customizing everything I can, and my MacBook setup here is no exception. So I want to show you guys a shortcut I found. My first thought was, can an AI product actually do this for me? And it turns out the answer is yes. A lot of people think Warp is an AI Vibe coding tool, but it's really more of a Vibe computering tool. There's so much you can do with it right in your computer terminal. For example, I can ask Warp's AI to remove all all the app icons in my dock that I haven't opened in the past 30 days. And I can actually see the AI thinking through the process inside my work terminal. And instead of having to go through each app myself, it just does it for me. I can also ask it to scan my home directory, find duplicate files older than 90 days, and move them into an archive folder. And I didn't have to go into my files once. And this, this is One of my favorites, Download all of the embedded images from this Google Doc in high resolution. If you've ever tried to download an image from a Google Doc before, you know why? This is actually a pretty killer shortcut. And there's so much more that you can do with Warp, both in terms of helping you code and also just helping you become the biggest power user of your own computer. And while other AI tools try to one shot, the task Warp keeps you in the loop so you and the agent can iterate together until it's exactly how you want it. It's a great example of a product that's not trying to replace developers, but just trying to make their lives easier. So if you're interested in an AI product that's built into your computer, try it out for free at the link in the description Even though Google pretty much dominated the news cycle this week, there was a lot more that happened. This week was also Microsoft's annual Ignite event, where they typically make a bunch of announcements. And well, this year's announcements were very focused on AI. One of the super interesting things to me that happened was that Microsoft, Nvidia and Anthropic all sort of teamed up together. And the reason this is interesting is because Microsoft has a huge percentage ownership in OpenAI, which is Anthropic's biggest competitor. Microsoft is committing up to 5 billion into anthropic, and anthropic is committed to purchasing 30 billion of Azure compute capacity. So despite Microsoft and OpenAI being super tied together, Microsoft is branching out into Anthropic now too. Now There were over 70 announcements at Microsoft Ignite, so I'm not going to get into all of them because many of them were geared towards enterprise. But but here's the ones that I found the most interesting. Microsoft is integrating AI agents directly into the Windows 11 taskbar. You'll be able to invoke Copilot and other agents from the taskbar or Start menu to automate or perform tasks on your PC. Get file summaries and more and you'll see the agent status bar in real time in your taskbar. You can interact with the floating windows and get notifications as agents work in the background. These Taskbar features, they're going to be opt in so you have control over whether or not you actually want to use them with your computer. Copilot is now going to be embedded in File Explorer, letting you summarize documents, answer questions, and draft emails with one click. We're also getting dedicated AI agents for Word, Excel and PowerPoint so you can create documents, spreadsheets and presentations from simple text prompts. You can then ask follow up questions to further tailor the content that it created. Now these aren't available yet for most consumers, but they claim they're coming soon. And also across the Copilot ecosystem, you're also getting access to the anthropic Claude models as a result of that partnership we just talked about. So in my opinion, those were the real standout features of Microsoft Ignite. But again, there was over 70 announcements. So if you work in like enterprise or you're a developer and you want to know specifically what they announced for you, make sure you check out the Future Tools IO news page because I shared all of the updates as they were coming out over on that website. On Monday of this week we got a new model out of xai with GRO 4.1 and for a moment it was pretty much the best model on the market. Gemini 3 came out the very next day, but Grok 4.1 was a pretty big upgrade from the previous Grok model. We can see that from the Elamarina tech's leaderboard. Grok 4.1 thinking pretty much won the day. And by day I mean like literally a day because the very next day Gemini 3 Pro came out and well, knocked it out of the lead. Apparently it's a leader in emotional intelligence, there's an EQ bench, and Grok 4.1 thinking is leading the pack there. In creative writing, it performed just slightly below GPT 5.1 and I think one of the real breakthroughs of this model was the much lower hallucination rate from previous models. Meta also released a couple of new models this week that are pretty dang impressive. Quite honestly, had it not been such an insane week with Google and Microsoft and everything else, these probably would have been like the top story of the week. So first one was Segment Anything Model three or SAM three. And this is a model where you can give it any sort of video or image and you can either click on a person or an object and it will highlight it, or you could type in something like people and it will highlight the people or highlight the fish in the image and then you can even separate them out so you can see three penguins and it found them all separately. So look at this image that they uploaded and look at all of the different things that it was able to segment out inside of this image. Like it was able to pick out every, every little thing, every fork, every bowl, every wine or champagne glass, even the lights on the Ceiling. It was able to pick each one of them out individually. If you go to AI demos.meta.com segment anything, you can actually play with this one as well. I had early access to this one, so I have spent a little bit of time with it already. But you can do some really cool stuff. Like, let me just pick one of their demo videos here. We've got this video of somebody playing soccer. Well, let's go ahead and search for an object. Let's search for soccer ball. It's searching for the soccer ball. It found it. Now let's tell it to search the entire video. It's going to scan the entire video and find the soccer ball throughout the entire thing. We can see it's following along. If you're a video editor, this type of thing saves so much time. Now that we've got the soccer ball tracked, I can continue to effects. And we can do things like put a contour around the soccer ball. And now it follows the soccer ball around with these giant lines. Let's put a glow on the soccer ball. Now we have a yellow glowing soccer ball. Let's make it oranges so it looks like the soccer ball is on fire. I could change the radius on the soccer ball as well. Well, and make it look just like a small orange outline or make it look like a big glow. Or I can magnify the soccer ball. Let's make the soccer ball bigger on the screen. Let's scale it up even more. Now we have a giant soccer ball that they're kicking around. I could also mess with everything but the soccer ball and change the background so that the background's blurred and the soccer ball is the only thing in focus. So really, really cool, fun stuff that you can do with video editing. But that wasn't the only model they released. They also released this one. Now, don't confuse this with SAM3. This one is SAM3D. Kind of the same idea where you can give it images and videos and select anything in it, and it will actually turn the thing you selected into a 3D object. So here's an image of a chair, and somebody selected that chair. And then they were able to visualize the 3D version of it in their room using an augmented reality app on their phone. Here's them highlighting an accordion and pulling just the accordion from this image. So I think I misspoke. I think I said it works with video and images. I'm thinking it's just images, but pretty impressive. I mean, look at all these things. They selected multiple objects. It went through SAM3D, did all of its, you know, AI stuff in the background and pulled all of those in as a 3D object. I'm imagining in the future, once this gets really good, you'll be able to take pictures of something, convert it to 3D, and then 3D, print versions of what you just took a picture of. And this one's also available to play with for free over in the segment. Anything Playground. You can create 3D scenes with SAM3D or create 3D bodies. Let's create. Try 3D scenes first. We have an image of some people sitting around, like coding, and I could just click on what I see in the scene. Let's look at this, like, little planter thing in the background. We can see it highlighted that. Let's also make sure it's got the plants coming out of it as well. And click generate 3D. And we can actually see it forming in 3D. And that actually happened pretty quickly. And look at how coherent that is from this image in the background here. Let's remove this stuff and let's get this dude and his chair. It's going to click until the whole thing's highlighted. And let's generate that. I mean, not perfect, but still pretty impressive, especially for how fast that was. Let's do create 3D bodies here. This one's probably a little bit better for people. Here's some people dancing. And it actually already found all the people, so. And I didn't actually have to prompt anything or click on anything. It just found the people for me. So let's select this dude here. And it's just automatically generating the 3D version. Oh, it's even got like the little skeleton inside, so definitely much better at getting people when you use this version of the model. I can also change this to a people reference view and see the 3D model among the rest of the people. Pretty, pretty, pretty cool. Like I said, this is probably something a lot more people would be talking about if Google hadn't dropped so much new stuff this week. Now for a quick recap of the AI drama from this week.

Foreign. Let's start with this super spicy tweet from Sam Altman. Hey, what the. But an actual drama. Larry Summers, one of the board members at OpenAI, decided to step down this week due to some dealings with some person named Epstein. I don't know what that's all about. All right, continuing on with some more OpenAI news, they actually released a new model this week called GPT 5.1 Codex Max. Yeah, they're still pretty good at naming things. This is their new Frontier agentic coding model and it's available inside of Codex. So this is more relevant if you're a coder, but it's a huge leap for coders because of the additional context length of this new model. This is their first model natively trained to operate across multiple context windows through a process called Compaction, coherently working over millions of tokens in a single task. So this unlocks project scale refactors, deep debugging sessions and multi hour agent loops so you can set their agent off on a task and let it just go do the work for like 24 hours. Compaction enables GPT 5.1 Codex Max to complete tasks that would have previously failed due to context window limits such as complex refactors and long running agent loops by pruning its history while preserving the most important context over long horizons. So if you're a chat GPT Plus Pro Business EDU Enterprise user, you have access to GPT 5.1 Codex Max. I haven't tested this model myself yet, but from what I'm hearing, it's a pretty, pretty dang impressive coding model. Hey, just had to jump in real quick because there's something I forgot to mention that OpenAI also rolled out. They actually rolled out a group chat feature. So when you're inside of your Chat GPT account, up in the top right, you'll see this little Start Group Chat button. If you click it and then select Start Group Chat, it'll give you an invite link that you can share with anybody. And now you can collaboratively chat inside of Chat GPT with multiple people in the same conversation. Here's what it looks like in practice. I went into ChatGPT and said give me 10 ideas for cool YouTube videos about AI. Gave me these. And then my producer Dave jumped in and said now incorporate Gary Busey into all of these ideas. And well, it did that. So a pretty cool new feature. I didn't want this video to go live without me mentioning it because I think a lot of people would be interested in it. But. All right, let me get back to the rest of the video. Now this is a cool little thing OpenAI is doing as well. There's a free version of Chat GPT built just for teachers. It's a secure ChatGPT workplace that supports teachers in their everyday work so they can focus on what matters. Teachers get free access to this through June of 2027. Here's what it comes with. Education grade security and compliance. So anything you Share with this model is not used to train their models and it's built to protect student data. It's got personalized teaching support so you can tell it to remember details like grade level, curriculum, preferred format so the responses are tailored to your teaching style. You can connect it to tools like Canva, Google Drive, Microsoft 365. It's got ready to use ideas and prompts with for teachers. In there you can create templates with other teachers and share them around. And it's got special admin controls for schools and district leaders. So if you're a teacher, I'd hit that up. And in the final bit of OpenAI news that I'm going to share today, Intuit signed a hundred million dollar deal with OpenAI to bring its apps to ChatGPT, which I actually think might be kind of cool. Like if all of your accounting software is connected to ChatGPT, maybe you'll just be able to say chatgpt, go do my taxes in the future. Get bent tax man.

That'll be fun. And I'm not done yet. There's a few more quick things that I need to share with you this week, so let's jump into a rapid fire.

Starting with the fact that Replit announced a new design feature. So if you're a Replit user, it actually now leverages Gemini 3.0 to help you create like really, really good designs. One of the problems with AI coding is if you're trying to build websites or apps, they all kind of look the same. I've heard people talk about how like the purple background and buttons is like the equivalent of the EM dash for coders. It's the dead giveaway that you used AI for coding and this new design feature actually creates like unique and good looking designs. Now, quick disclaimer. I do have a little bit of an investment in replit, but I would have shared the news regardless because I do think it actually makes pretty cool looking design designs. In an interesting move, eleven Labs introduced image and video into their platform. With eleven Labs, you can now bring ideas to life in one complete creative workflow. Use models like Vo Sora, Cling Juan and Seed Dance to create high quality visuals. Then bring them to life with the best voices, music and sound effects from eleven Labs. So eleven Labs is kind of pulling in all the features from all of the platforms and is almost trying to be like what what Leonardo and Crea and Higgs Field and some of these other platforms are. It seems like eleven Labs wants to be a one stop shop for everything creative AI, but in most people's minds they're pretty much cemented as like the audio generation platform. Manus this week introduced the Manus Browser operator. This is an extension that works on any of the Chromium browsers, including of course, Chrome. You install this extension and it can do agentic things right inside of your browser, take control of your screen, click around for you, do searches for you, book restaurants for you. A lot of the stuff that we've seen a lot of these other agentic browsers do. Well, now you can do it with the Manus browser operator extension inside of Chrome as well. Also, Mid Journey is back in the news this week. Now that's a name I've not heard in a long time. They just rolled out a new profiles feature. I haven't talked about Mid Journey in a long time, so I wanted to point this out to show you that I am still paying attention to Mid Journey. I just haven't felt like a lot of the announcements they've made recently have been that that exciting or exciting enough to share. But this one's actually a pretty cool upgrade. They're sort of building in community features inside of Mid Journey where you can have your own profile and people can view the images you generated and you can share your various social media platforms. My profile is midjourney.com mistereflow as you can see, I haven't finished doing anything with it like setting profile images or banner images, but I did put my social media links there. Now I've got one last thing I want to share with you that I really think you're going to enjoy. But before I do, I do want to confess that I detoured a little bit on this channel from sharing all of the AI news. However, I quickly realized that I absolutely love staying informed on the AI news. Reading the articles, talking to the people, playing with the tools, demoing stuff like that's what I live for. I absolutely love it. And I pretty quickly started to miss sharing the news. So I finished my detour. I'm back on track. I'm gonna start releasing these weekly news videos again and giving you the breakdowns as well as making individual news videos for the big event like I did this week. I'm getting this channel back on track, doing the thing that probably got you to follow my channel in the first place. I apologize for the detour, but I just like sharing the news too much. It's too fun to me. I don't know why I even decided to pause it in the first place. With that being said, I want to end this video with something that I came across online that is now probably my favorite video on the Internet. I introduced to you Russia unveiling their first humanoid autonomous robot.

Thank you so much for nerding out with me today. If you like videos like this, make sure to give it a thumbs up and subscribe to this channel. Channel. I'll make sure more videos like this show up in your YouTube feed. And if you haven't already, check out FutureTools IO, where I share all the coolest AI tools and all the latest AI news. And there's an awesome free newsletter. Thanks again. Really appreciate you. See you in the next one.