## Intro [00:00] Google has released Gemini CLI and while it feels like a knockoff of Cloud Code, it does have a lot of powerful features. So, in this video, I'm going to look at the top five features in the Gemini CLI and then we'll look at how it codes and compare it with Claude Code. Let's go. ## Free tier [00:15] Gemini CLI has a free tier that's very generous. Like, check out some of the things you get. You get access to the Gemini 2.5 Pro model, which is one of the best models that Google has for AI coding. I think actually the is the best model right now. You get 60 requests per minute and a thousand requests per day at no charge. You can do that all with your Google account and you don't have to put in a credit card. You just have to have a standard Google account. Like this is pretty crazy and compared to something like Claude Code where the base plan is 20 bucks a month and that you don't even get very much. So this is going to be a real game changer for people. We'll see how long it lasts. But for right now, this is a crazy opportunity to use leading AI models totally for free. ## Google Search [00:55] If there's one thing Google knows how to do really well is web search and they even fuse that into the Gemini CLI. You see in the documentation here it says ground your queries with the Google search tool built into Gemini. The way that works is it all the prompts you give it if it doesn't have all the information it needs or it needs to search for more current information on the internet. It has access to the full Google search platform to do those searches and bring that information back. Now we're in the Gemini CLI. Now, how it looks there is you're going to see sometimes it's going to have a Google search tool here. Then once it gets back the information, it's going to ground the results with that information it's getting from its Google search. This is actually really powerful because it gives you latest information on technologies using the best standards for coding practices. You just get it out of the box. It just integrates so nicely with Gemini. ## NonInteractive Mode [01:42] You don't just have to use the Gemini CLI in interactive mode. You can also use it anywhere in your terminal to do any basically anything really. All you have to do is just after you put in Gemini, just do a d-prompt and that's going to say run it in non-interactive mode. So we can just say who won the 2025 Super Bowl. There we go. It just gives us a little bit of information about the the Eagles big win. This is really powerful because you can pipe in files as input and also take the output and pipe that to a file as well. So you can do all kinds of different scripts and workflows with this feature. So in this more useful example, I'm going to say summarize this project and bullet points. And now it's going to take the context of the current project it's in and just give me a write up and markdown of all the different things. So I can just pipe this to a markdown file and a piece of documentation automatically generated for me. So Gemini CLI can really superpower your entire terminal. ## Context Window [02:30] One really amazing feature with the Gemini CLI is how big the context window you get is. It's 1 million tokens. So here what I'm doing is I'm loading in an entire project. is actually the entire Shopify site and I'm loading it into the context and we can see here the context is going down but like this is a huge site and it's just going down by 20% 30% and there we go it load the whole thing and I still have 64% of the context left. So what this means is now that I'm doing changes to this site have the entire site as context. So I can say it's going to know when something changes over on file A that affects file B. It's going to know all the different relationships and if I make a change somewhere, it's not going to break something else. So the huge context, the 1 million token context is a huge feature of Gemini. ## Multimodal [03:20] Gemini CLI is fully multimodal. So check this out. I can just take this invoice file, just say convert to ajson file. So basically extract the text from it and then turn it into a nice JSON file that I can use. In just a few seconds, it created invoice.json. Here's all the data put in there. And I haven't tried this myself, but apparently you can use the VO MCP servers and actually create images and little videos right from inside Gemini CLI. So you can really do everything from input to output, all kinds of multimodal, from video, images, sound. It's actually very powerful in that respect. If you're into AI software development like me, make sure you subscribe to my newsletter, the AI unleashed news. It's the first link in the description, and I hope to see you there. So, while all those features are amazing, if it coding abilities isn't up to par with something like cloud code or cursor, then it's all useless. So, I'm going to take a real example that I'm working on right now and see how Gemini CLI handles it versus cloud code. This just a simple application I've written and what it does is allows a Shopify site to show customers on a map where the retail locations are. So, if you click on one of these markers, it shows right now the street view for the location. What I want to add to it is a directions button. So you can click on that. It'll go right to the Google's direction and navigate you to the to the store. We're going to say in store locator add a directions button to the location marker so the user can get Google Maps directions to the location. So the first thing that's going to have to do is figure out in the codebase of the Shopify application exactly where that map store locator is. So it looks like it's done a good job to to search through that and find it. This is where having that huge context is going to help a lot where it can load up a whole bunch of fi different files and then be able to quickly find what it needs to change. So, I have it accepting edits. And one thing I just noticed here, it added these non-English characters in the file and I took them out again, which I guess is okay, but I never seen stuff like that in claw code. Bit of a weird hallucination there. Another thing you'll notice is down in the bottom right here, it's actually switched me to Gemini 2.5 Flash. I've seen this happen a few times now where you start off on Gemini 2.5 Pro and then after some coding it's going to switch you to the inferior flash model. But just watch out for that. Okay, so now it's finished and I said I've added the directions button to the locations marker in the store locator which will open Google Maps with directions to the selected location. All right, well it sounds good. Let's see how it works. Now if we jump back into our map, we just click on one of these locations. Okay, so now we see the directions button. Click on that. It opens up a new tab. It puts my location there and then tells me exactly how to get to this location that I clicked on. So, no complaints here. It was a simple task, but Gemini CLI nailed it. See how it would handle something a little bit more challenging. So, actually once I restarted Gemini, now it's back to Gemini 2.5 Pro. That's another tip I've kind of discovered. If you get switched to Flash, try to restart it and get back to Pro cuz you really want the Pro one. Let's say update the default zoom based on the user's device resolution so they can see the same viewport. Definitely a more tricky ask cuz now it's going to ask to figure out what resolution the user has and then adjust the the zoom on the map to so they see the same number of locations. If we see here it's taken a pretty simplistic approach. It's it's getting the width of the window that the the map's in. And then if it's less than 768 pixels has a a zoom of 7. And then if it's less than 1200 a zoom of 8. And if it's bigger than that a zoom of 9. Let's see if this approach to the solution is actually the right one. if it works well and then we'll see if cla code can do even better. Now, if we look at what it actually did, it actually made it worse in a lot of ways. So, if we just open it up normally, it's going to zoom in too much on the Vancouver area. It's not going to expand enough to show the other locations. But then, if we switch to our phone view and refresh that now, this looks a little bit better, but it's still not wide enough to get a lot of the locations over here. So, what it really should have done is tried to be smarter about it and try to figure out what viewport it should use to get the most locations on the different devices. I'd call this one a decent attempt but a definitely a fail. You see with clawed code using exactly the same prompt, exactly the same code base, it had a much better implementation. What I did is it calculated the effective width. So it took the screen width times the pixel ratio. Then it had more categories of effective width with different zoom rates. This should work much better. And the result of that cloud code change seems to be much more effective than what Gemini CLI did. So now on desktop mode it shows a pretty good representation of the locations. I switch to the mobile refresh that pretty good as well. I think that's actually a pretty good representation of the differences between where Gemini CLI is and where Claude Code is. For simple things, Gemini CLI can keep up just fine. Once you get to more complex features and ask it to do more complex things, then Claude Code and the Cloud 4 models still have a big edge up. Let me know in the comments what you think. If you had a chance to try the Gemini CLI, make sure you subscribe to the channel if you like this video. Hope you're having an amazing day.