Antigravity 2.0 Just Changed AI Coding Forever! ft Gemini 3.5 Flash

Añadido: 2026-05-23

[00:00:00]So, Google just unveiled Gemini 3.5 Flash, anti-gravity 2.0, and a new anti-gravity CLI, and even a anti-gravity ID just yesterday. And in this video, we're going to take a look at the newest version of anti-gravity and see what changed and how to use the new anti-gravity 2.0. And the thing is that quite a lot has changed with anti-gravity. Now, with anti-gravity 2.0, you don't get to use the ID or let's say the cord editor directly within the platform. Instead, you will have to download and install a separate Google anti-gravity ID package like a entirely different software. And not just that, they also unveiled anti-gravity CLI. And it's finally time to say goodbye to Gemini CLI. Okay, so there's a lot to talk in today's video.

[00:00:43]And first of all, we'll begin with Gemini 3.5 Flash. And this model is interesting for a couple of reason.

[00:00:49]First of all, it says Gemini 3.5 is built to help you execute complex and agentic workflows. Now the interesting thing here is that see always flash models were inferior to pro models right so obviously Gemini 3 pro will always be superior than Gemini 3 flash but things have turned around quite a lot in here now Gemini 3.5 flash is able to deliver a far superior performance when compared with 3.1 pro that is to say Gemini 3.5 flash which is actually less expensive than Gemini 3.1 Pro is actually outperforming 3.1 Pro in a lot of agentic task and you know a lot of benchmarks And not just that, it is even outperforming GPD 5.5 and even clude OPUS 4.7. And as you can see here, we have the benchmark charts. And maybe I can zoom in a little bit. And as you can see here, we have Gemini 3.5 Flash. And it is being compared to Gemini 3 Flash, Gemini 3.1 Pro, Cloud Sonet 4.6, Opus 4.7, and even GPD 5.5. Now take a look at some of these uh interesting benchmark results. Terminal Bench 2.1, SW Bench Pro, MCP Atlas, Tool, Finance Agent, etc. Now first of all we'll compare it with Gemini 3.1 Pro and in almost all of them as you can see Gemini 3.5 flash is outperforming Gemini 3.1 Pro directly in almost all of these I mean in the most of the benchmark test Gemini 3.5 flash is topping the chart and even when compared with Clude Oppus 4.7 as you can see in certain benchmark results like uh terminal bench 2.1 Gemini 3.5 flash is actually outperforming cloud opus 4.7 so Gemini flash is supposed to be like a lighter version of the model like far inferior than let's say Gemini 3.1 Pro or even the solid series but right now with the newest version claude oppus 4.7 is being dethroned by Gemini 3.5 flash now the thing is that in this particular blog post I was not able to find any pricing information regarding this 3.5 you know uh flash model okay so that felt interesting because when open anthropic or even Gemini for that matter introduces a new model within this blog post itself they usually provide their pricing information then I went to the Gemini developer API pricing section right here and here I found the Gemini 3.5 flash pricing. Okay. And as you can see the input price for per million input token is $1.5 and output price is actually $9. And at first it might not sound like a lot right because Gemini 3.5 flash is able to deliver superior performance when compared with Cloud Opus 4.7 4.6 and GPT 5.5 in certain benchmarks and these models are already costing more. For example, Claude Opus 4.6 7 is $5 per million input tokens and about $25 per million output. And again for Open AIGT 5.5, it is like $5 input and about $30 output. But the thing is that if you take a look at the pricing history of the previous flash models, for example, maybe let's just say let's take 2.0 Flash. As you can see, Gemini 2.0 Flash was just 10 cents for input and about 40 cents for output. Okay, per million of course. And then let's take a look at the Gemini 2.5 flash. Now the price bumped up that is $0.30 per million input token and $25 per million output token. Next up if we take a look at let's say Gemini 3 flash as you can see $0.5 per million input token and $3 per million output token. And now when we click on Gemini 3.5 flash as you can see input price is at $1.5 but output price have seen like a significant jump that is $9 per million output token. So, Gemini 3.5 flash is getting really expensive. Of course, it is a capable model for long horizon agentic task and all that, but pricing is something that you should keep an eye on and I just found it interesting and I wanted to share it right away. And next up, let's just explore anti-gravity 2.0 and the new anti-gravity ID. But before that, let me quickly take a second to talk about the sponsors for today's video, Porkburn. If you're building an online store, selling digital products, doing drop shipping, or even running a side hustle, one thing that matters more than people think is your domain name. Right now, you can grab a dot shop domain for just $1 for the first year. So, instead of using some long random URL, you can get a clean and brandable domain, something like your brand.shop, which instantly tells people that you're selling something online. And one thing I like a lot about Porkbun is that they keep things simple. No aggressive upsells, no hidden renewals or prices, and they even include free who is privacy, free SSL certificates, email forwarding, and a bunch of features that other registars usually charge extra for. The dashboard is super straightforward, and even if this is your first time purchasing a domain, they do have real human support if you ever get stuck. Plus, PBIN has been ranked as one of the top domain registars by Forbes Advisor and USA Today. So, they have built a pretty solid reputation. So, if you want to launch your store or secure your brand name, just click the link in the description below to get your doshop domain today for just $1 for the first year. Now, back to the video. And now, if you want to start using Andy Gravity 2.0, you can directly head over to Andygravity.google and click on this download button. And as you can see here, we have the option for downloading Andygravity 2.0 for Mac OS, Windows or Linux. And again, if you already have Andyravity installed, you can update it to the latest version. So, I did the same. And here I have opened up and this is Andy Gravity 2.0. And a lot has changed in here. First of all, now it is actually moving to a agent first approach similar to cursor cordex and all that. And now you don't get a integrated development environment that is an ID or let's say editor interface inside of Andy gravity 2.0. Yes, you heard that correct. Even within cursor you get to open the legacy editor mode and nowiggravity ID is a separate standalone software that you'll have to download and install. And when you open Andy gravity 2.0 this is going to be the kind of interface that you'll see.

[00:06:26]Towards the left side you can find all these projects and also the option for creating a new conversation conversation history and also scheduling some task and towards the right side here we have the option for giving the prompt selecting the project and even selecting a model and nowhere in here you find the option for actually invoking or let's say opening up the editor and again maybe I can just open up a sample conversation so I just opened up a blank folder and asked AI to set up a fresh nextjs project in this folder and as you can see here it have went ahead and done everything. Now the thing is that you only get to use Gemini models or whatever models that are currently available within anti-gravity inside of antigravity 2.0. That is to say earlier if you wanted to let's say use cloud code inside of anti-gravity like running parallel workflows or let's say working on the same project handing off between anti-gravity and cloud code within the same interface. Well, that's not now possible because right now you don't get a terminal within this uh and gravity 2.0. Of course, as you can see, npm rundev, which is like a command to start the dev server is running. And if I click on the same towards the right side, it opens up this one right here.

[00:07:30]But as you can see, I don't find an option to let's say manually type in a command or let's say create a new terminal instance or anything as such.

[00:07:38]Everything is handled by anti-gravity itself. And again here we have review, diff mode and all that. Now the interesting thing is that every time you give a prompt, anti-gravity 2.0 do will go ahead and sprawl like sub agents and you'll be able to find those sub agents in here. For example, maybe I can let's say open a new conversation and I can say can you please create a PDF merge split image to PDF website and I'll select the 3.5 flash high model and I can click on the send button and as you can see it says working so let's just wait for it and as you can see it says I'll start by checking the current d and it says I'll start by checking the current content of the workspace directory to understand if you're working with an existing project or starting completely from scratch. Okay, so let's just see. So there you go.

[00:08:25]Andigraphy 2.0 have come up with a implementation plan and I can click on it. Okay, looks good. I can click on proceed and if I click on overview let's see if it actually uses sub agents. If that's the case then I'll be able to show them in here. Okay, seems like the agent is not using sub aents. Actually a single agent is doing the work. In which case, if the AA was able to let's say use sub agents or let's say multiple agents to work on the same task, you would have been able to see those sub agents in here and you could have clicked on the sub agent in here and found more information. So that is basically the main change in here and as you can see right now the AI is making several changes to our code and there's no way to directly view the entire files and folders within this project. After making all these changes, you'll be able to review the code in diff mode but that's about it. Let's say you want to access the editor and view all these files and take it forward. In that case, you'll have to head back to Andygravity.

[00:09:18]Google and if you scroll all the way down, you will find a new section called as Andy gravity IDE. Now, this is like a separate standalone package that you'll have to download and install on Mac OS, Windows or Linux. And after that, you can open it up. So, here I have opened up Andy Gravity ID. So, I'll quickly go ahead and log into my Google account.

[00:09:34]Okay. Click on next. Start fresh. Okay.

[00:09:38]I can just fast forward everything in here. There you go. Here we have anti-gravity ID. So if you are about to use let's say anti-gravity 2.0, you don't get access to all these extensions, cord editor or anything as such. And for that you will have to get the anti-gravity IDE. And I don't know what I feel about this. But it would have been great if both of them were integrated into the same package. And also giving direct terminal access is huge. But sadly that is not available in Andy Gravity 2.0. And I kind of have like a mixed feeling about it. And now now that we have like Andy gravity CLI that's an entirely different thing and I'll talk about it in a later video. But as you saw Andygravity 2.0 is actually like a agent first interface where you don't get to access the terminal directly. you don't get to see any files or let's say folders directly or you don't even get a editor and for that you'll have to download and install a separate package and every time the AI makes some changes towards the right side you'll be able to find all the files to which the AI made changes and you can click on it directly and then view the code and this is all about it and I still couldn't find any places where I can actually view the entire files or let's say folders within the project and now yet another new feature right inside of Andy Gravity 2.0 is scheduled task. As you can see here, we have a schedule task option. So after opening a project, you can click on schedule task and click on this new option right here. And now you can give it a name and select a project and select a schedule and give a prompt. And now what happens is that Andygravity 2.0 itself will go ahead and invoke this particular schedule task when the schedule time is reached. And again that's also new scheduling feature right inside of Andy Gravity 2.0. And if you open ID, this is like the OG anti-gravity that we are used to, right?

[00:11:14]So towards the right side here we have the agent panel and you can invoke agent ask it to do things you can review the files but at the same time you also have the option for viewing all these files in here directly make changes use tab edits and do all that kind of stuff and this is like a integrated experience and not just that if you let's say want to use cloud code directly within this interface itself in that case you can actually go ahead and open a new terminal and write the cloud command hit enter and you can directly start using it right inside within anti-gravity itself right and now it is not possible for you to do it directly within Andy Gravity 2.0. In that case, you'll have to open a separate terminal instance or let's say Andygravity ID itself and then start using it in there. In which case, here we have the output I believe. Okay, I can click on submit and let's just wait for it. So, I just wanted to show you how to use Andyravity 2.0 and how it is actually fundamentally different from Andygravity that we are all used to. And here I'll open cursor. So, as you can see here, I have opened cursor the latest version of cursor and the thing is that it is also agent first. So you also have the option for selecting a project and creating a new agent, a cloud one and a local one. But the interesting thing here is that in just a single click you'll be able to open the editor window. So towards the top as you can see here we have the editor window option and as soon as I click on the same it opens up a ID like interface where I can open the folder. Okay for example let's just say okay let's just open any of these folder and here we have the normal ID kind of interface.

[00:12:38]And if I want to switch back to the agents window, I can click on this agents window option and it goes back.

[00:12:43]And not just that, it also gives us the option to let's say have multiple instances of terminal. So I can simply click on this icon right here, open the side panel and then I can invoke any number of uh let's say terminal and maybe if I want to use cloud, I can hit enter like type and hit enter and I can directly start using it. And not just that within this agent first interface itself. If you click on this icon right here I'll be able to find all the code folders files everything else within this project as well. So it would have been like nice additions in gravity 2.0.

[00:13:13]And to be honest I actually missed them a lot in Andy 2.0. So these are some of the fundamental changes in Andy 2.0. And again if you're looking for that OG editor interface now you'll have to download and install a separate package and you don't get to use other tools like for example let's say cloud code within Andy gravity 2.0. And for that you'll have to open terminal or let's say open terminal inside of antiigravity ID. And before I wind up the video, let's take a look at Testbrite 3.2 and also their new web portal. So Testbrite is a autonomous A powered software testing agent that can do a complete front end and backend test of your W coded app. And to get started, all you got to do is to click the link in the description below or head over to testprite.com. And as you can see, it says 10x your A native dev speed with our autonomous AI testing agent handling verification. So first of all I'll quickly go ahead and sign into my account and as for using test sprite you primarily have like two options and the first one is MCP test and the second one is web test and for MCP test you simply go to the MCP test section click on the quick install button add this MCP to cursor or whatever a coding agent that you're using for example and gravity and then after adding the MCP you can give a prompt and run the test directly from your ID but now you also have the option for running the test directly on the platform itself and that too on the deployed version or let's say a staged ution of your app. Okay. So, first of all, I'll click on this create test option. And this time around, I'll select this option right here that says live web app. And I can click on continue. And next step, let's just say imag. And now I need to upload a product PRD. Now, this right here is like a sample app that I created using Andy Gravity 2.0. And I can go ahead and log to my account real quick. So, it is basically like a AI image generation platform that I created in just a single prompt just to show you a demo. So let me quickly login and I can give a prompt like a cat generate image. There you go. Here we have the image of the cat and also if I go to history section I'll be able to find all the images that I' have generated within the platform. Okay. So this is basically the functionality of the app. And next up what I'll do is that I'll head back to test sprite and first of all I'll start by uploading the product PRD. So for this one I actually created a PRD using anti-gravity itself and I'll select the same in here and I can click on next and it is actually reading the documentation and find all the use cases and test cases. So basically you just provide the PRD and the A itself will go ahead and find all the use cases and features and as you can see it says we found four features and nine use cases from the documentation and again if you zoom in you'll be able to find more information. Okay and next up I need to enter the API base URL. If I'm about to do let's say the back end test, I need to enter all this API information in here. Since we are doing the front end, I'll go ahead and copy the front end URL. Paste it in here. Okay. So, we have already provided the uh PRD. I can enter the username and password and I can click on next.

[00:16:05]Okay, there you go. As you can see, it says exploring your application. So, our agent is visiting each feature from your product map to understand how the app behaves. This usually takes about 3 minutes. So basically what happened is that we have provided the PRD and also the user login information along with our app address. Now the AI agent within test project will go through our entire app and map out all the pages all the features and find what all need to be tested and what kind of test need to be carried out and prepare like an extensive list of everything that needs to be done and as you can see in the screen itself you can find a live screen recording of the A agent actually exploring our app. As you can see it is trying to like register for a new account and then it will go ahead and do everything. So let's just wait for it.

[00:16:44]All right. So there you go. Test sprite explored our entire app, found all the features and generated all these test plan details in here and it have planned like 19 different test and now all I got to do is to click on this generate test option right here. And as you can see here we have all these 19 test mapped out and it will go ahead and start doing each of them one by one. So here we have sign out from the history page, view generation history order, return to the app after authentication, protected page, redirect missing sessions to login, sign in with invalid credentials, generate an image from a prompt, and here we have a bunch of different test that the AI is planning to carry out.

[00:17:18]And it will take some time for the AI to actually carry out and complete all these test. And all you got to do is to sit back and wait. And after it's done, we can go ahead and explore the test results and we'll take it up from there.

[00:17:29]All right. So there you go. Out of 19 test 18 of them is complete and one of them is still in progress. And out of this only two of the test failed and two of them got blocked and currently one is running. For example, I can actually go ahead and let's say uh take a look at one of them. For example, let's just say view image generation history order. So I can click on the same and now it opens up a detailed popover that shows every details about this particular test. And towards the right side I can find a screen recording of the A actually performing the test. As you can see, it filled in all the credentials and logged into our account. And there you go. That is actually working. And if you want to, you can actually go ahead and regenerate this test by giving or let's say adjusting the uh steps in here. And next up, let's just go ahead and open this one right here. That is generate an image from a prompt. Hm, that's interesting because it is supposed to work. So, I can click on play. And let's see. So, the agent is trying to login. Okay, there you go. It logged in. And next up, it entered a prompt. select an aspect ratio and okay clicked on the generate image button and it actually showed unauthorized. So that is basically like a error that we have and we need to fix that. Okay, got it. And as you can see here we have the complete details of all the tests that the AI carried out. And if you click on the site exploration that is AI agent right here you'll be able to find all these features that the AI was able to find. And again if you click on this one it opens up the A agent actually navigating around our website and doing the thing that is exploring the platform and making use of this feature that is generating a image.

[00:19:02]Let's see. As you can see the A actually went ahead and filled in a prompt and generated an image. And here we have the output. And if I click on agent actions I'll be able to find all these things that the AI agent have actually done.

[00:19:13]And once the entire testing process is complete it will go ahead and prepare like a test report. And as you can see this right here is like an extensive test report which contains all the information about all the test that was carried out. What were the errors, what to fix and how to go about it. And yeah, this is how you basically use Testbrite 3.0 and their new web test feature right inside of Testbrite. And again, if you want to check that out, just head over to testprite.com and sign up for a new account and you can instantly start testing your application. And yeah, that's pretty much all I wanted to show you in this section. So this is everything that I wanted to talk about in today's video. So basically we have Gemini 3.5 flash A model, anti-gravity 2.0, anti-gravity ID and also anti-gravity CLI. And I'll be making a dedicated video about anti-gravity CLI soon. So make sure you're subscribed for that and I'll see you in the next one.

[00:19:57]Bye.