Go back

Gemini 4 Neon is too good and too cheap

16m 6s

Gemini 4 Neon is too good and too cheap

This video reviews Gemini 4 Neon, a new Google model that has generated significant online buzz for being fast, cheap, and highly capable. The host, sponsored by Google but promising an honest take, argues the model is a major leap, with some users comparing it to Opus 5.5. He showcases examples from the internet, including motion design for a pelican mascot, a robotics dog stress test with per-paw slope adaptation, a physics-rich pelican riding a bike that outperforms GPT 6.2 Astra, and playable pinball games built with Blender and WebGPU. A key insight is that Neon writes all this in raw code (HTML, SVG, three.js) in one shot rather than acting as a video or motion design tool, demonstrating strong coding and spatial reasoning. The host explains where to use the model: the Gemini app for quick visual prototypes, AI Studio for large multimodal context up to one million tokens, and Anti-Gravity for app building, where performance reportedly jumps from 12.8% to 39%. He shares best practices such as setting thinking to high, using the orchestrator for daily work and swarms for large builds, adding verification loops, and putting context first. He then suggests monetization ideas: playable lead magnets and 3D configurators, custom project demos as sales pitches, financial simulators, back-office automation, and overnight micro-products. Finally, he offers honest caveats: Neon can be overly brief, exhibits AI writing tics like em dashes and "not X, it's Y" phrasing, has confusing naming and tier gating with no free tier, and Anti-Gravity still has room for improvement. Overall, he concludes Neon is an insane, cost-effective model worth exploring.

Transcription

2721 Words, 14337 Characters

English
Speaker 1Is Gemini 4 Neon worth your time? The internet is going wild for it. They just can't believe how good of a model it is. It's fast, but the most important thing is it's cheap. What could you do if you had twice the amount of agents building your ideas? Would you be more productive? Would you make more money? This is the question. Some people are saying it's as good as Opus 5.5. Today we're going to go find out. I'm going to show you wild examples on the internet of people using neon and what they're able to create in vibe manufacturing and video games, etc. We're going to talk about how you can get the most of Gemini 4 Neon, and we're going to talk about ideas to make money with it, be productive with it, just to get your creative juices flowing. What does this unlock? So this is the short and sweet guide to Gemini 4 Neon, why you should care. Thanks to Google for sponsoring today's episode, but I'm going to keep it real. I'm going to keep it real. I'm going to say what I like about it, what I don't like about it, to share with a friend, like, comment, and subscribe for more of this sort of stuff in your feed. And this is just the de facto guide to it. TLDR, by the way, this is an insane model because it's cost effective and you can do a lot with it. And I'm excited for what you can build with it. Let's get into it. Let's talk about some examples before we get into some ideas. being more productive and also how you can actually some best practices with using neon so some so uh chet chet hasula said uh here's gemini form neon motion design for a pelican product i mean when you look at the motion design here it looks pretty darn good and you know i'm one of those people that i'm building apps i'm looking for mascots i'm looking for motion design so something that can do this really well um just looks this looks really awesome let's keep moving google launch neon i've been testing it inside anti-gravity we'll talk about that later why you should be building an anti-gravity um versus some of the other platforms um but it's really strong at coding i'm still testing it with a few more things to share but here's something i built with it hope you enjoy when you look at this it is in it is just well done it is you know it feels really like i'm looking at this and like it it is almost unbelievable that we're at a stage where google model is doing this where ai is doing this um so when i'm looking at this i'm thinking about what are hardware companies I can create, what are prototypes I can be creating. I did a whole episode on Argonne before where I talked about Vibe Manufacturing and stuff like that. So I think Gemini 4 Neon just feels like it does a really good job at some of this hardware stuff, Vibe Manufacturing stuff. Let's keep it moving. We're going to keep it quick. This episode is everything you need to know about it. Okay, so you all wanted me to push Gemini for Neon further, so I did. I asked it to add uneven limestone terrain to stress test the robotics dog's biomechanics. It implemented per-paw slope adaptation, five-point obstacle scanning, dynamics, spine torsion, all this stuff. Look at this. You know, it reminds me of the Boston Dynamics dogs. I'm looking at this and it's, again, hardware, movement, you know, gaming. You can imagine games, worlds that you can create. This looks stunning. I'm curious what some of the replies say. Gemini going up, damn. Yeah, I think a lot of us, someone said, well, I'm actually impressed. I definitely need to try out this model. I think a lot of people have basically been sleeping on a lot of these Gemini models for a while and now it feels like Google is back. Let's keep them moving. I really like this one. So on the left, we're going to show the left and right. Initially, this is Gemini 4 Neon versus GPT 6.2 Astra. This is Gemini 4 Neon. Made a pelican riding a bike for three minutes. I think Google's UI design is a bit better than ChatGPT's with the solid sense of aesthetic. The model is so darn good. I mean, look at this versus GPT-6 to Astra. I mean, from a physics perspective, you're just not getting that high quality. Look how high quality the physics is. This is like a different world. I mean, yes, the design, there's some birds here. We got some birds here too. but you know so there's some debate around what is nicer um but overall i think just from a hardcore physics perspective and i bet that if you actually had some you know reference images um you can probably get add some of these things here as well and just to be clear what's wild about this pelican animation and some of the other demos i'm gonna show you isn't that neon is spitting out a video file or acting like a motion design tool it's literally writing all of this in raw code so it's html it's svg it's 3gs and it's in one shot and that's why it's such a crazy flex of its coding and spatial reasoning anyways physics awesome um this is someone who created two pinball games using blender web gpu optimized and generated play field art derived pbr stained glass maps and multiball physics with procedural uh audio so this is crazy this is a literal pinball game um looks amazing kind of reminds me of like the wind actually this really reminds me of like the windows xp uh pinball game i know you all um this is feels like that so uh this is crazy it's live people can play it they're playing it you can go and ship this stuff okay so those are some examples of some crazy things you can do motion graphics physics cad vibe manufacturing what do you need to know i'm going to go through this really quick a lot of people are going to wonder do i use this on the gemini app do i use this on ai studio do i use this on anti-gravity for gemini you can use it in all three but here's here's where and why you should use it so for the gemini app you want to use it for like these fast 60 second visual prototypes playable lead magnet magnets for workspace ops like live formula sheets gmails and docs for ai studio you've got that full 1 million uh token context so that's for like that multimodal huge multimodal dumps like audio video csvs it also shows the live token count and anti-gravity caps at 512k so that's really uh key um and then for anti-gravity that's where you're actually building your apps it's where you're building big things so this is what some other folks are talking about around there's a three times coding lift um 12.8 to 39 plus a built-in browser qa on anti-gravity now the built-in browser in my opinion anti-gravity isn't particularly great but you have something you also have it as an orchestrator so you have neon and brand new hydrogen daily and you can do swarms for big builds we're going to talk about swarms in a second but swarms are a really good way to have multiple agents working in parallel getting things done so um the i'm going to talk about the best practices uh for getting things out of neon and then i'm going to talk about how you can make money be more productive with it some ideas to get your creative juices flowing um so five ways to get the most out of neon um set thinking this high I've noticed that high is just like the sweet spot. So it's a sweet spot for speed and cost. It's 32% cheaper than carbon. Extra high burns 70% more for less than two points. It's just, in my opinion, not worth it. Do as you wish. But, you know, we don't want to, we don't waste. Friends don't let friends waste tokens. Build an anti-gravity, not the bear wrapper. So the same model jumped 12.8% to 39% inside anti-gravity. Some people were talking smack on X, being like, oh, it's really not that good of a model. And then other people were like, it's insane. It's as good as 5.5. One of the reasons why is because some people were using an anti-gravity, some people weren't. So if you're building, like we talked about, building big things, you're working on big projects, just do it in anti-gravity. Three, use the orchestrator for daily work and swarms for big builds. Swarms, like I talked about, this is how you can have sub-agents do. a particular work, you do slash swarm, and then you could say, have a researcher do this, have a front-end designer do this, have a front-end engineer do this, have a back-end engineer. This is you spinning up your AI team to go and build things. This is going to help you get more for big builds. And this is something that you can do before you go to sleep and you wake up in the morning and you have a big project done. Always give it a verification loop. So tell to open the browser, click buttons, and fix bugs. Always surprised that some people don't do this. You can do this for Neon, you can do this for any of the Google models, or you can use it for any AI model. Something you should be thinking about. And lastly, put context first and demand facts. Triggers, automatic caching, and stop sales, coffee, drift. That's to get the most out of it. All right, so what are some ways think about how you can make money, how you can be more productive with Neon. This is just going to get your creative juices flowing. With Neon, you can actually create playable lead magnets and 3D configurators, mini games. So these are things that capture emails that get shared on X or on Instagram. Agencies would cost $10,000, $25,000, $50,000, $100,000 for these things that you can now, you saw how good the 3D was. You saw how good the physics were. You saw that you can do motion graphics. How do you create these 3GS customizers that capture emails? You create these 3D apps. No brainer to be doing that as a way to get attention for whatever it is you're building. Number two, Trojan horse custom project demos. So I tried this in my testings. I had early testing. You can put in a website URL and then, it'll just go and scrape things from that website or go, so for example, I don't know, Chick-fil-A, yeah, Chick-fil-A, Toronto. It'll go, so assuming that was Chick-fil-A slash Toronto, it would go and search that website. It would grab all that information about it. And then you can actually build something for that store, like a calculator, a tool, something that's valuable to you. And just go and send it to them. You've built something like this. I think that there's so much money to be made for pitches that actually have something. And of course, it might not be, and I'm going to do, I should do a whole episode on this, because I think that if I was like, especially if I was like 20 years old, trying to make cash flow, one of the things I would do would be build websites for people and send them the websites. Like, look at, look for companies that have kind of like stale-looking websites. Redesign it then. Be like, hey, I have this website for you. Would you pay $100 a month? Would you pay $50 a month to have it? Oh, what don't you like about it? I'll edit it for you. These are things that you can do. Trojan horse custom project demos. Three, live formula margin and pricing simulators. So you can take messy invoices in. Neon is so good with data. And you get a working financial model and audit out. I can't share but I share my screen on this because it's got sensitive data. But I took my own company's data, looked at it, and now I'm like going back and forth. I feel like I've got this like CFO-like person helping me out. Four, 24-7 back office automation. So you can do things like lead routing with Neon. You can do CRM ops. You can do reconciliation. And it's just a lot cheaper than some of the competition out there. And five, overnight micro-products and directories. So with Swarm, you can have agent teams researching, building, and QA and just testing things while you ship. So these are just some ways to get the creative juices flowing. And there's obviously more, but just like for uniquely Neon things, these are some things to think about. But before we head out, I need to give you some just four honest caveats, keeping it real, about Google Gemini for Neon. One is that I've noticed that Neon can be overly brief. Sometimes, I've actually compared it to Fable 5.2 versus Neon. What I'll get from Fable 5.2 is just this long, long, long answer. Then I go to Neon, and it's like 33% of the length. Now sometimes that's really good. Sometimes I don't want a lot. But sometimes I do want a lot. So you have to ask it if you want a lot, if you're used to just sort of longer answers to things, you might be surprised with how brief it is. Two, I don't know if it's just me, but I've just noticed em dashes galore. I've also noticed when writing, they talk about it's not X, it's Y. So a lot of the AI smell in terms of writing is there. That being said, the actual writing itself is amazing from my testing so far. So I will say that if you tell me that you're using Neon, then I'm going to tell you that Neon is amazing. But if you tell it not to do em dashes and you tell it not to do X, not Y, which are just like AI smell writing things, you probably will get some really good output there. So I think it's low-key a really good writer. Three, can we talk about how confusing the naming and the tier gating is just with Google products? Hydrogen, Neon, Carbon, you've got 512k versus a million. There's no free tier. So it's a bit confusing. But once you get past it, you're cooking. And then lastly, I talked about anti-gravity has some stuff to be desired. Google's moving so quick right now in terms of the models. You've got to think that anti-gravity team is probably adding features soon. But as of right now, as of recording this, November 19, 2026, anti-gravity has some stuff to be desired. Hope this has been helpful. Do you want me to do more of this stuff? When new tools come out, when new models come out, do you want me to give you my quick take? Or do you just want me to stick to ideas, growth tactics? I like showing you things that I'm using, things that, you know, how I would get the most out of it. But please do let me know. I live to serve. I want you to get the most out of AI, to build out your ideas, to ship your greatest things. And I read every single YouTube comment. I respond to a lot, too. And if you want me to create more videos, like, comment, and subscribe. Thank you so much. Happy building. Have a creative day. And I'll see you next time on the Startup Ideas Podcast.

Podcast Summary

Key Points:

  1. Gemini 4 Neon is praised as a fast, cost-effective model that the internet is excited about, with some comparing it to Opus 5.5.
  2. The model generates impressive motion design, physics simulations, and 3D content by writing raw code (HTML, SVG, three.js) in one shot rather than producing video files.
  3. Examples shown include a pelican motion design, a robotics dog stress test, a pelican riding a bike compared favorably to GPT 6.2 Astra, and playable pinball games built with Blender and WebGPU.
  4. Users can access Neon through the Gemini app for quick prototypes, AI Studio for large multimodal context (1 million tokens), and Anti-Gravity for building apps, where it reportedly gains a 12.8% to 39% coding lift.
  5. Best practices include setting thinking to high, building in Anti-Gravity, using the orchestrator for daily work and swarms for big builds, adding verification loops, and putting context first.
  6. Money-making ideas include playable lead magnets, 3D configurators, custom project demos as pitches, financial and pricing simulators, back-office automation, and overnight micro-products.
  7. Caveats include overly brief responses, AI-style writing tics like em dashes and "not X, it's Y" phrasing, confusing naming and tier gating with no free tier, and Anti-Gravity still needing improvement.

Summary:

This video reviews Gemini 4 Neon, a new Google model that has generated significant online buzz for being fast, cheap, and highly capable. The host, sponsored by Google but promising an honest take, argues the model is a major leap, with some users comparing it to Opus 5.5. He showcases examples from the internet, including motion design for a pelican mascot, a robotics dog stress test with per-paw slope adaptation, a physics-rich pelican riding a bike that outperforms GPT 6.2 Astra, and playable pinball games built with Blender and WebGPU. A key insight is that Neon writes all this in raw code (HTML, SVG, three.js) in one shot rather than acting as a video or motion design tool, demonstrating strong coding and spatial reasoning.

The host explains where to use the model: the Gemini app for quick visual prototypes, AI Studio for large multimodal context up to one million tokens, and Anti-Gravity for app building, where performance reportedly jumps from 12.8% to 39%. He shares best practices such as setting thinking to high, using the orchestrator for daily work and swarms for large builds, adding verification loops, and putting context first. He then suggests monetization ideas: playable lead magnets and 3D configurators, custom project demos as sales pitches, financial simulators, back-office automation, and overnight micro-products. Finally, he offers honest caveats: Neon can be overly brief, exhibits AI writing tics like em dashes and "not X, it's Y" phrasing, has confusing naming and tier gating with no free tier, and Anti-Gravity still has room for improvement. Overall, he concludes Neon is an insane, cost-effective model worth exploring.

FAQs

It is praised for being fast, cost effective, and strong at coding, physics, and spatial reasoning. It can generate motion design, 3D scenes, and games in raw code like HTML, SVG, and Three.js.

Use the Gemini app for quick 60-second visual prototypes, AI Studio for large multimodal inputs with its 1 million token context, and Anti-Gravity for building larger apps.

High is the recommended sweet spot for speed and cost. Extra high burns about 70% more tokens for less than two points of improvement, so it is generally not worth it.

Swarms let you spin up multiple sub-agents, such as researchers, designers, and engineers, to work in parallel. Use them for big builds, for example by running them overnight to finish a large project.

You can build playable lead magnets, 3D configurators, mini games, custom project demos for pitching clients, pricing simulators, back-office automations, and overnight micro-products or directories.

It can be overly brief compared to other models, and its writing often includes AI-sounding habits like em dashes and "not X, it's Y" phrasing. The naming and tier gating across Google products is also confusing, and there is no free tier.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.