Codex vs Gemini CLI vs Claude Code: Pricing and Usability Compared

Three CLIs, Three Philosophies

The AI coding assistant has officially moved from the IDE into the terminal. OpenAI, Google, and Anthropic each ship a flagship command-line tool that turns their best model into a hands-on agent that reads files, runs commands, and edits your code. They look superficially similar but feel very different in practice. This article compares the three on the two dimensions that matter most when you adopt one for daily use: pricing and usability.

Codex CLI (OpenAI)

Codex CLI is OpenAI's open-source terminal agent, written largely in Rust, designed to run as a sandboxed coding assistant on your machine. It ships under the Apache 2.0 license and pairs naturally with the GPT-5 family of models.

Pricing. If you already pay for a ChatGPT Plus ($20/mo), Business, or Pro ($200/mo) plan, you can sign in to Codex CLI with that account and your usage is bundled in, with quotas that scale with the tier. If you prefer the API, GPT-5 sits at roughly $1.25 per million input tokens and $10 per million output tokens, which is among the cheapest frontier-class options. There is also a small free API credit when you start.

Usability. Installation is one npm command. The TUI is clean but a little spartan compared to Claude Code. Codex runs commands in a sandbox by default, which is a strong safety story; you opt in to broader permissions when you need them. The big lever is the model: when GPT-5 with high reasoning effort is the right tool, Codex CLI gives you direct access to it from your shell.

Gemini CLI (Google)

Gemini CLI is Google's open-source terminal agent, also Apache 2.0 licensed, built around the Gemini 2.5 family. Its standout feature is a free tier that is unusually generous by industry standards.

Pricing. Sign in with a personal Google account and you get free use of Gemini 2.5 Pro with quotas of roughly 60 model requests per minute and 1,000 per day at no charge. That is enough to do real work without ever paying. If you outgrow it, Gemini Code Assist Standard is around $19 per user per month and the Enterprise tier is around $45. On the API, Gemini 2.5 Pro is priced similarly to GPT-5 at about $1.25 in / $10 out per million tokens, with cheaper rates for shorter prompts.

Usability. The CLI is straightforward and supports Model Context Protocol servers, multimodal inputs, and a 1 million token context window which is the largest of the three. Where it lags is the agent loop itself: Gemini's tool-use planning and self-correction are noticeably less mature than Claude Code's, and the underlying model still trails the others on hard coding benchmarks. Great for cost-sensitive work; less great when you need it to grind through a complex multi-file refactor unattended.

Claude Code (Anthropic)

Claude Code is Anthropic's terminal-native coding agent, paired with the Claude Sonnet and Opus models. It is widely regarded as the most capable of the three at end-to-end coding tasks.

Pricing. A Claude Pro plan ($20/mo) includes light Claude Code usage. Serious users almost always step up to Claude Max at $100/mo (about 5x usage) or $200/mo (about 20x usage), which is where the platform shines for daily-driver work. If you go API-only, Sonnet runs about $3 per million input and $15 per million output tokens, while Opus is roughly $15 / $75. That is the most expensive of the three at the API level, and heavy agent loops can burn tokens fast.

Usability. Claude Code has the most polished agent experience of the trio. The TUI feels considered, slash commands and sub-agents are first-class, hooks let you wire in custom lifecycle behavior, MCP integration is excellent, and a project-level CLAUDE.md file gives you persistent context that survives session resets. The extended thinking escalation (think, think hard, ultrathink), parallel sub-agents, and plan mode are unique productivity multipliers. The price for all this is that you will, eventually, hit usage limits or token costs that the others would not have triggered.

Pricing Side-by-Side

For a developer who codes a few hours a day, the cheapest path is Gemini CLI on its free tier. If you already pay $20/mo for ChatGPT, Codex CLI is effectively free incremental tooling. If you want the best coding agent and you write code for a living, the Claude Max tier at $100 to $200 per month is the natural home, and it will usually still come out cheaper than running equivalent work through the raw API. At the API level, GPT-5 and Gemini 2.5 Pro are roughly tied on cost, while Claude Sonnet is 2 to 3x more expensive per token but often delivers the work in fewer iterations.

Usability Side-by-Side

Claude Code wins on agent maturity, ergonomics, and the depth of its extensibility story. Codex CLI is the safety-first option with the cleanest sandboxing model and an actively improving experience now that GPT-5 is the underlying engine. Gemini CLI is the most forgiving for newcomers because the free tier removes the meter anxiety, and the 1M context window is genuinely useful when you point it at a large repository.

Which One Should You Pick?

Pick Gemini CLI if cost matters most and your work is bounded enough that the free quotas cover it. Pick Codex CLI if you already pay for ChatGPT or you want a sandboxed, open-source agent backed by GPT-5. Pick Claude Code if your time is more expensive than your tokens and you want the agent that most reliably ships finished work. Many engineers run two of them side by side: Gemini or Codex for cheap, broad exploration, and Claude Code for the hard execution. There is no rule against using all three.