The Rise of AI CLIs: A Deep Dive into Gemini, Claude, and OpenAI

Gemini, Claude, and OpenAI have all released their own terminal-based CLIs to integrate with AI agents. And just recently, another one, Quinn Code, has joined the growing list of coding assistants. This raises several key questions that this article aims to answer: What is the advantage of working in a terminal? Do these specialized terminal CLIs offer AI models that outperform standard ones like Claude or Gemini? There's a noticeable trend of integrations, with tools like Cursor now embedding Claude's capabilities directly into the IDE for easier access.

Many developers have just grown accustomed to "vibe coding" with a persistent AI chat in their IDE. The idea of closing a familiar coding editor to work exclusively in the terminal might seem daunting. Yet, the developer community buzz suggests this is the direction the industry is heading. So, let's examine Gemini CLI, Claude Code, and OpenAI's Codex CLI to see what they offer.

Exploring Google's Gemini CLI

Gemini CLI is an open-source tool available for download on GitHub for both Windows and macOS. You can execute it directly using npx or install it globally via npm.

# Run with NPX
npx gemini

# Or install globally
npm install -g gemini-cli

A significant advantage is its generous free tier, which provides access to Gemini 2.5 Pro. This powerful model allows for up to 60 requests per minute and a total of 1,000 requests per day, all at no cost.

After installation, running gemini in the terminal initiates the setup process, which includes selecting a theme and logging into a Google account.

Editing and Animating a Website

To test its capabilities, the first task is to edit a website and add animations to the hero section. The website in question is built on Wix Studio and uses a proprietary programming language called Velo, which presents a learning curve. While the platform includes a built-in coding terminal, connecting the site to a GitHub repository allows for local development in a familiar environment like VS Code.

Inside VS Code, with the Gemini terminal open, a request was made to add an animation to the homepage. Much like a standard AI chat, it processes the request and makes the necessary changes. However, a key difference is the ability to run multiple instances of Gemini in parallel, allowing several tasks to be completed simultaneously in the background.

After Gemini finished its work and the changes were pushed to production, a preview of the website revealed a new fade-in animation that triggers one second after loading.

Advanced Integrations and Concurrent Operations

Gemini CLI also features the ability to connect to MCP (Master Control Program) servers. By using a JSON configuration file, Gemini can be relaunched to authenticate with a Wix MCP server, granting it access to make changes on the user's behalf.

{
  "server": "wix-mcp-server-url",
  "auth_token": "your-auth-token"
}

The power of this tool becomes even more apparent when running multiple Gemini instances concurrently. With its large context window, it's possible to push its limits significantly. As a test, Gemini was tasked with creating a complete financial calculator for budgeting from scratch. It generated a high-level project plan, built the application using Next.js, and delivered a functional result.

While more polished one-shot solutions might be found on platforms like Vercel or Bolt, this exercise doesn't fully capture the strengths of a CLI-based workflow. With the project now running, the large context window and multiple CLI instances can be leveraged to install numerous features simultaneously.

Gemini CLI also offers other functionalities, such as web searching and diagram creation, but it's wise to monitor usage to avoid exhausting the free tier credits prematurely.

A Look at Anthropic's Claude Code

Claude Code is considered by some to be the original AI CLI, predating the recent surge in popularity of similar tools. While Gemini gained traction quickly, Claude Code has maintained steady growth due to its reliable and consistent performance.

Installation is similar to Gemini CLI; it's installed as a global package via the terminal.

npm install -g claude-code

After installation, you need to provide billing details by connecting it to the Anthropic API or a paid Claude account.

Getting started with Claude Code feels more structured. Its initial recommendation is to create a README.md file by running the /initialize command. This process scans the project, using some tokens to build a foundational context and understanding of the codebase. The resulting documentation is so well-written that it would be valuable for any engineer joining the project.

Building a Financial App with Claude Code

To compare its performance with Gemini, Claude Code was tasked with building a similar financial application. With Claude Code integrated into the Cursor IDE, a prompt similar to the one used before was provided: build a budgeting app from scratch.

After several weeks of use, a distinct pattern emerged in how Claude Code operates compared to other AI tools, especially in a chat-based interface. It tends to use a significant number of tokens to meticulously plan its course of action. This isn't a drawback; observing its process of creating a detailed plan, executing it, and then thoroughly testing the results reveals why it performs so effectively.

The resulting budgeting app was visually superior to the one created by Gemini CLI. It included more features, such as adding expenses categorized by week, month, or year, and calculating monthly savings based on the established budget.

A Note on Version Control

A notable drawback when using tools like Claude Code or Gemini is the lack of built-in version control. The revision history available in IDEs like Cursor or VS Code, which allows for reviewing, approving, or rejecting changes, is absent. When Claude Code makes changes, there is no visual diff to track them.

A workaround for this is to rely more heavily on Git. By committing changes frequently, it's possible to track the modifications made by the AI and discard them if they are not satisfactory. It is hoped that future versions of these coding CLIs will incorporate better version tracking, but for now, this method is effective.

Claude instances can also run simultaneously. This can be achieved by opening multiple terminal windows for different feature requests or by instructing Claude to run sub-agents in parallel, which can be configured in the system prompt. Be mindful that running numerous requests at once can consume a large number of tokens.

OpenAI's Codex CLI

The final contender is OpenAI's Codex CLI. It is also available on GitHub for terminal installation, but it uniquely offers a web version that allows for direct interaction with a GitHub repository, providing a more user-friendly way to manage parallel tasks.

The Web Version Experience

Trying the web version first, it connects to a GitHub account. Opening the same Wix Studio site project from earlier, it automatically creates three initial tasks: explain the codebase, check for bugs, and identify potential improvements.

Confirming these tasks sets them all running in parallel in the background, each with its own agent. Expanding a task reveals a container starting up and a log of its activity, including terminal snapshots as it progresses.

For the first task, it provided a summary of the project, correctly identifying it as a Wix Studio site and outlining the codebase structure. For bug fixing, it identified a more efficient way to write the command for hiding and showing the top menu. The interface provides an option to create a pull request directly from the findings, which can then be reviewed and merged into the repository.

The Terminal Experience

After installing the CLI version and logging into an OpenAI account, the codex command can be run from within a project directory.

Tasked with creating the same financial budgeting app, it ran in the background for a similar amount of time as Gemini and Claude. However, unlike the other two, it did not automatically install Next.js. This required opening a new terminal tab to manually install dependencies and run the project.

The final result was functionally equivalent to the others but lacked the visual polish of the Gemini and Claude versions. It also missed some of the advanced features from the Claude-generated app, such as weekly, monthly, and yearly expense tracking.

Final Thoughts

After a thorough comparison, Claude Code stands out as the most impressive tool. Its structured approach, reliable performance, and high-quality output make it the top recommendation for anyone looking to integrate an AI CLI into their development workflow.