On June 25th, Google released the open-source Gemini CLI, bringing its most powerful Gemini 2.5 model straight into the terminal.

Meanwhile, Claude Code by Anthropic has become a go-to option for developers who like working in the terminal. Both tools help you write, fix, and manage code without switching to an editor. But how do they handle real coding tasks?

In this blog, we'll test both on practical problems, from small games to tricky bug fixes, to see how they think, where they struggle, and which one you'd actually want to use.

TL;DR

Claude Code and Gemini CLI are both strong tools for coding from the terminal, but they focus on different things.

Claude Code: Feels more like a developer assistant. It asks smart follow-up questions, gives a cleaner UI, and sometimes builds more than you asked, like full games with effects and animations.

Gemini CLI: Focuses more on utility. It works fast, sticks to the prompt, and is great for quick fixes or shell-level tasks, but the output often feels a bit rushed or too minimal.

Brief on Claude Code

Claude Code is a terminal-based tool by Anthropic that helps you write, read, fix, and commit code, all from the command line. You install it with a single npm command, and start working by typing claude in your terminal.

It opens up a chat-like prompt where you can ask it to do things like:

Claude reads your files directly from the project folder, no need to point it at specific files. It understands the structure and logic pretty well and suggests changes that usually make sense. It won't rush into edits either. Before doing anything, it asks for your approval. That makes it feel safe to use.

You can also use it for Git operations:

Claude Code remembers things across sessions, too. It uses memory files like CLAUDE.md to keep track of your past commands, explanations, or custom instructions. You can manage this memory with simple commands like /memory to see what it knows, or /compact to shrink it down when needed.

For more advanced setups, Claude Code supports MCP (Model Context Protocol). This lets it connect to other tools on your machine, like test runners, databases, or even local APIs. That means you can ask it things like "run the tests" or "query users from local DB" and it can actually do it, no tabs switching.

Claude Code supports /help, session resume (claude -c), and even quick one-liners with claude -p. If you're already a Claude Pro or Max user, you can start using it right away without any extra cost.

It's fast, clean, and works well for most everyday dev tasks, especially if you're already used to working in a terminal.

Brief on Gemini CLI

Gemini CLI is Google's command-line tool that brings its Gemini models directly into your terminal. It works like a smart coding assistant that can read your codebase, help you write and edit files, run shell commands, and automate workflows, all inside your terminal window.

You can start using it instantly with this:

Or install it globally:

The first time you run it, you'll be asked to sign in with your Google account. That gives you access to 1,000 free requests a day and up to 60 requests per minute. If you want more control or want to choose your model, you can also add a Gemini API key from Google AI Studio.

What makes Gemini CLI stand out is its support for large codebases and complex tasks. It can read multiple files, generate apps from PDFs or screenshots, and even summarize Git history or build reports, just by asking in plain language. It also supports media generation, slide deck creation, and local tool integration through something called MCP (Model Context Protocol), which lets it talk to external tools like file systems or image generators.

Here's the kind of stuff you can ask it:

You don't need to configure much. Just run the command in your project folder and start typing. It keeps your current files and folders in scope, so you don't have to re-explain things. It feels like talking to someone who's already looking at your code.

What We're Comparing

Claude Code and Gemini CLI both run in your terminal and help with things like editing code, fixing bugs, and managing Git, all without leaving the command line.

But they go about it differently.

So instead of listing features, we'll give them the same real coding tasks and see how each one handles it, how they think, what they miss, and where they shine.

Task 1: Build a Flip Card Memory Game

To test both tools in a quick coding scenario, we asked them to build a flip card memory game. This is the prompt we used:

Claude Code:

Claude Code started by asking a few clarifying questions, like "Do you want images, emojis on the cards, or just colors?" Once I confirmed the requirements, it planned the full setup, one HTML file with all the logic inside. This is what Claude comes up with.

The UI looks great, the functionalities are working, and it properly uses the emojis, as I've mentioned.

Code:

Gemini CLI:

I used the same prompt with Gemini CLI, and it came up with this result:

The UI looks completely wrong. There is no styling, and all the cards are in a straight line, making everything look bad.

Code:

Task 2: Fixing the Drag & Drop Reorder List

We gave both tools a buggy version of a task list app. Items could be dragged, but dropping was inconsistent and often broke if you didn't land exactly on a list item.

This is the code that requires fixing.

I gave this prompt to each one:

Gemini CLI:

To test with Gemini, we just need to tag that file using @ sign and it will be able to read and update that file. This is what Gemini came up with.

Gemini kept it simple and reliable. It fixed all drop position issues using getDragAfterElement, added clean animations, and gave the interface a nice modern touch. The logic is clear, short, and easy to follow. It's exactly the kind of fix you'd expect from a developer trying to patch something quickly but cleanly.

Code:

Claude Code:

Claude also followed the same approach. We mentioned the file first, and then it is able to understand and is ready to make changes. It came up with this result.

Claude went a few steps further. It not only fixed the bugs but also built a mini drag-and-drop framework with smooth transitions, shadows, and even touch support. It added a button to generate random tasks with emojis and reorder them. It's a fuller UI/UX revamp, not just a patch.

In fact, this type of behavior, going beyond just "fixing" and truly understanding the UX intent, is exactly what Entelligence aims to do at scale. If you work with larger codebases or want to track the development of features week by week, Entelligence can provide insights at the repository level, generate comprehensive documentation, and identify architectural patterns without requiring you to explain everything from scratch.

Task 3: Click-to-Shoot in 3D Space

We gave both tools a slightly more advanced challenge: create a simple shooting game in 3D space.

We've used this prompt:

Gemini CLI:

Gemini built a working 3D scene with static shapes using Three.js. Clicking on them changed their color, but that's where the interaction ended. There was no animation, no feedback, and no sense of a real game, just clickable shapes sitting still in space.

Code:

Claude Code:

We also gave the same prompt to Claude, and it came up with this result.

We can shoot the enemies, which generates confetti, and they disappear. After shooting them all, an alert shows up. It's good, but it still needs a lot of improvement.

Code:

Pricing & Usage

Gemini CLI is free to use with your Google account. You get up to 1,000 requests per day and 60 requests per minute, which is more than enough for most coding work. You can also plug in your own Gemini API key from Google AI Studio if you want more control or need higher limits.

Claude Code is available for Claude Pro and Claude Max users. That means if you already pay for Anthropic's Claude service, you don't need to pay anything extra, Claude Code is included. There's no free tier, so if you're not a subscriber, you won't be able to use it.

So in short:

Conclusion

Both Gemini CLI and Claude Code are helpful in their own ways.

Gemini CLI is better if you just want to fix something quickly, run a few commands, or get short answers without much setup. It stays out of your way and works well for small tasks or command-line routines.

Claude Code, on the other hand, is more thoughtful. It handles bigger tasks, understands your code better, and even helps with design, Git, and building things from scratch. It feels more like a coding partner than a tool.

In the end, it depends on how you work. If you like to move fast with simple prompts, Gemini is solid. But if you're doing deeper work and want better support across the board, Claude Code gives you more to work with.

Links:

• Claude Code Overview – Anthropic Docs

• Claude Code Setup & Installation Guide

• Gemini CLI GitHub

• Gemini CLI Setup & Installation Guide