How much do these CLI coding agents cost?

Claude Code is usage-based via the Anthropic API, with Claude Pro from $20/month and Max plans (e.g. $100/month) for predictable billing. Codex CLI is included with ChatGPT Plus (~$20/month); its 1M context may need the $200 Pro plan. Gemini CLI is free with generous limits, though Pro token budgets have tightened.

Which CLI agent has the best sandboxing?

Codex CLI has the strongest sandboxing — OS-level isolation via macOS Seatbelt and Linux Landlock enforced at the kernel, plus three-tier permissions (read-only, auto, full-auto). Claude Code uses namespace-based isolation with network blocking. Cursor added OS-level sandboxing in early 2026 but is IDE-based, not a CLI.

Quick Answer

Claude Code vs Codex CLI vs Gemini CLI 2026: Which to Pick

Q: What is the best CLI coding agent in 2026?

Claude Code is best for deep multi-file reasoning and large refactors (now on Opus 5/Sonnet 5, 1M context, hook/plugin system). Codex CLI is best for secure, sandboxed execution and background tasks (OS-level Seatbelt/Landlock sandboxing, open-source Rust). Gemini CLI is best for large codebases and Google Cloud users, with a generous free tier.

Published: August 2, 2026

The Short Answer

For CLI coding agents in 2026: Claude Code for deep reasoning and big refactors, Codex CLI for secure sandboxed execution and background tasks, Gemini CLI for large codebases and Google Cloud users with a generous free tier.

The Comparison

	Claude Code	Codex CLI	Gemini CLI
Models	Opus 5 / Sonnet 5	GPT-5.6 Sol / Codex	Gemini 3.5/3.6
Context	1M	Up to 1M (Pro)	1M+
Sandboxing	Namespace + net block	OS-level (Seatbelt/Landlock)	Ecosystem-specific
Pricing	$20+/mo (usage)	ChatGPT Plus ~$20/mo	Free (generous)
Best at	Reasoning, refactors	Secure exec, background tasks	Large repos, Google Cloud

Where Each Wins

Claude Code → reasoning depth. Best at unfamiliar codebases, multi-file tasks, and large refactors. Full hook/plugin system and a /loop command for recurring lint/test monitoring. Ideal for Anthropic-standardized teams.
Codex CLI → security + delegation. OS-level sandboxing at the kernel, three-tier permissions, open-source (Apache 2.0, Rust). Great for parallel background tasks and generating PRs without tying up your machine. By May 2026, 70%+ of users delegated tasks estimated at 1+ hour.
Gemini CLI → scale + free tier. 1M+ context, strong for large repos and analysis, tight Google Cloud integration. Best cost-sensitive option, though the free tier often defaults to lighter Flash models.

How To Choose

Complex refactors / unfamiliar code → Claude Code.
Kernel-enforced safety / background PRs → Codex CLI.
Huge repos / Google Cloud / free → Gemini CLI.

Many of these now support SKILL.md for portable skills across agents.

Verdict

Deepest reasoning → Claude Code
Most secure execution → Codex CLI
Best for large repos + free → Gemini CLI

Sources

Anthropic — Claude Code: anthropic.com/claude-code
OpenAI Codex: openai.com/index/how-agents-are-transforming-work
Google — Gemini CLI docs: ai.google.dev/gemini-api/docs