Gemini CLI vs Claude Code vs Codex CLI: Three Free Terminal Agents Compared (2026)

Between 2025 and 2026, all three major tech giants released their own terminal AI coding agents: Google’s Gemini CLI, Anthropic’s Claude Code, and OpenAI’s Codex CLI. All three run in the terminal, but they differ significantly in pricing models, free quotas, and capability focus.

At a Glance

Dimension	Gemini CLI	Claude Code	Codex CLI
Free Quota	1,000 requests/day (personal account)	No standalone free tier	Requires ChatGPT Plus
Minimum Monthly Cost	$0	$0 (pay-as-you-go with API Key)	$20 (ChatGPT Plus)
Core Model	Gemini 2.5 Pro	Claude Opus 4.6	o3 / o4-mini
Context Window	1 million tokens	200K tokens	Standard
Open Source	Yes	Yes	Yes
Available in China	No	No	No
MCP Support	Yes	Yes	Yes

---

Free Quota Breakdown

Gemini CLI: The Most Generous Free Tier

Sign in with a personal Google account to get:

• 60 requests/minute
• 1,000 requests/day
• Free access to Gemini 2.5 Pro (a top-tier model worth hundreds of dollars)

For the vast majority of developers, this is more than enough for daily use — no credit card required.

Claude Code: Pay-as-You-Go, No Fixed Free Quota

Claude Code is not tied to a subscription; it uses Anthropic API Keys directly:

• No fixed free quota for individual use (new accounts receive a small amount of free credits)
• A Max subscription ($100-200/month) is available for high-usage guarantees
• Maximum flexibility — pay only for what you use, no waste

Codex CLI: Requires ChatGPT Plus

• Requires ChatGPT Plus ($20/month) or pay-as-you-go with an OpenAI API Key
• The Codex quota included with a Plus subscription is capped (approximately 30-150 tasks per 5-hour window)
• The API Key approach is flexible, but o3 model pricing is relatively high

Zero-cost recommendation: Gemini CLI. A personal Google account gets you 1,000 requests/day, far outpacing the other two.

---

Model Capabilities Compared

Gemini CLI — Gemini 2.5 Pro

• 1-million-token ultra-long context window, capable of processing an entire large codebase in one shot
• Built-in Google Search Grounding for real-time access to the latest documentation and APIs
• Multimodal support (image understanding)

Claude Code — Claude Opus 4.6

• Widely regarded as producing top-tier code quality, especially excelling in complex reasoning and architecture design
• CLAUDE.md project specification mechanism for deep project context understanding
• The most comprehensive Git workflow integration (commits, PRs, branch management)

Codex CLI — o3 / o4-mini

• OpenAI reasoning models with advantages in algorithmic and mathematical tasks
• Full-screen TUI interface offering the most IDE-like interactive experience
• MCP support for extensible external tools

---

Installation

# Gemini CLI
npm install -g @google/gemini-cli
gemini  # Sign in with your Google account

# Claude Code
npm install -g @anthropic-ai/claude-code
claude  # Requires ANTHROPIC_API_KEY

# Codex CLI
npm install -g @openai/codex
# or
brew install --cask codex
codex  # Requires OPENAI_API_KEY or ChatGPT login

---

Use Case Comparison

Scenario	Recommendation
Zero-cost access to top-tier models	Gemini CLI
Processing very large codebases (>100K lines)	Gemini CLI (1M token context)
Complex architecture design / code refactoring	Claude Code
Git workflow automation	Claude Code
Algorithm/math-intensive tasks	Codex CLI (o3 reasoning)
Already subscribed to ChatGPT Plus	Codex CLI (no additional cost)

---

Usage in China

All three require access to overseas APIs and cannot be used directly from within China:

• Gemini CLI -> Google API (blocked)
• Claude Code -> Anthropic API (blocked)
• Codex CLI -> OpenAI API (blocked)

Developers in China may consider Cline + Volcengine Ark Coding Plan (setup guide) as an alternative.

---

Final Recommendations

• Want free access: Choose Gemini CLI — 1,000 requests/day is more than enough
• Best code quality: Choose Claude Code — unmatched for architecture and refactoring
• Already a ChatGPT Plus subscriber: Choose Codex CLI — included at no extra cost
• Very large codebases: Gemini CLI’s 1-million-token context window is irreplaceable