Claude Code vs Codex vs Gemini CLI

As of 22 October 2026, Claude Code has the deepest agent harness and the most creator workflows built on it, but it starts at $20 a month, while Codex is the only one of the three you can start free and its CLI is open source. Gemini CLI stopped serving free personal Google accounts on 18 June 2026 [26], so it now fits only teams that already pay for Gemini Code Assist or a Gemini API key.

Last verified 2026-10-22 · prices and features from vendor docs; usage from our corpus of saved posts

What changed in 2026What creators actually use each one forWhich one to pick

Verified 2026-10-22 from vendor docs, pricing pages and the official GitHub repos. Prices in USD per month.
Claude CodeCodexGemini CLI
Maker and CLI licenseAnthropic. Proprietary, all rights reserved [4]OpenAI. Apache-2.0 [18]Google. Apache-2.0 [27]
Free way inNone. The Claude Free plan does not include Claude Code [1]ChatGPT Free, for quick coding tasks; OpenAI publishes no numeric limit [15]None for personal Google accounts since 18 June 2026; Google moved them to Antigravity CLI [26]
Cheapest paid way inClaude Pro, $20 ($17 billed yearly) [1]ChatGPT Go $8 or Plus $20 [15]A paid Gemini API key or Vertex key, or a Gemini Code Assist Standard license [26]
Heavier useMax from $100 (5x) and $200 (20x); Team $25 or $125 per seat billed monthly; Enterprise $20 per seat plus usage at API rates [1] [14]Pro from $100 ($100, $200, $500); Business $25 per user billed monthly [15]Code Assist Standard 1,500 requests per day, Enterprise 2,000 per day [28]
Pay per token insteadAnthropic API key, or Amazon Bedrock, Google Cloud and Microsoft Foundry [2]OpenAI API key, billed at API rates [15]Gemini API key or Vertex AI [27]
Default model and contextOpus 5.5 by default; Fable 5.1 is opt-in. Current models get a 1M-token context window [3]GPT-6.1 Sol recommended for coding. Context window not published in the Codex docs [16]Auto routing across Gemini 3 and 2.5 models [30]. Gemini 4 Argon has a 1M-token context, and the README points Argon users to Antigravity instead [31] [27]
Lifecycle hooksYes [5]Yes [19]Yes [33]
Agent Skills (SKILL.md)Yes [6]Yes [20]Yes [34]
Plugins or extensionsPlugins and marketplaces [7]Plugins [20]Extensions [35]
MCP serversYes [8]Yes [21]Yes [36]
SubagentsYes [9]Yes [22]Yes [37]
Parallel workGit worktrees, background agents, cloud sessions on the web [10]Git worktrees plus Codex Cloud tasks [23]Git worktrees, marked experimental; no cloud tasks [38]
Headless runsclaude -p [11]codex exec [24]gemini -p with JSON output [39]
Project memory fileCLAUDE.md, and it also reads AGENTS.md [12]AGENTS.md [25]GEMINI.md [40]
Where it runsTerminal, VS Code and Cursor, JetBrains (beta), desktop app, web, mobile [2]Terminal, VS Code, Cursor and Windsurf, JetBrains, Xcode, desktop app, web, mobile [17]Terminal, VS Code companion extension, JetBrains via the ACP agent registry; no own app [41]
Latest release (checked 2026-10-22)v2.1.309, 22 Oct 2026 [4]rust-v0.163.0, 15 Oct 2026 [18]v0.65.0, 20 Oct 2026 [27]

What changed in 2026

What creators actually use each one for

Claude Code posts are mostly about extending the harness rather than raw coding. Boris Cherny's team tips (Claude Code Team Tips — Boris Cherny's 10 habits for parallel…) and his CLAUDE.md (Boris Cherny's personal CLAUDE.md — plan-first, self-improving rules) set the plan-first pattern, and roundups like 7 must-install Claude Code repos — Obsidian Skills, n8n-MCP… list the skill packs people install. A second cluster is about stretching usage limits: rerouting the harness through OpenRouter (Claude Code on OpenRouter — keep the harness, swap in cheaper models) or letting the top model plan while cheaper subagents execute (Fable orchestrates, Opus executes — cutting frontier-model token spend).

Codex shows up most as a second opinion on Claude's work. Creators run it as an auditor of Claude Code output (Codex + Claude Code — 3 systems to audit CC output with Codex), cross-audit plans between the two models (Claudex skill — Claude and GPT-6 Astra cross-audit plan and build), or call it from inside Claude Code with OpenAI's plugin (Fable 5 plus GPT-5.6 routing — Codex plugin inside Claude Code). Solo builds lean on its scheduled tasks (GPT-6 Astra + Codex one-person company — context files and daily…) and its hands-off asset work, such as a Blender model built and embedded with Three.js (Codex + Blender — agent-built 3D hero asset embedded with Three.js). One tier list ranks Codex and Claude Code level and calls Codex cheaper for comparable output (AI tool tier list May 2026 — Codex and Claude Code top, chatbots C).

Gemini CLI has one strong entry, and it is now historical: a post about the free 1,000-requests-a-day tier (Gemini CLI — free terminal coding agent, 1K req/day, Gemini 2.5 Pro). Creators who wanted a free terminal agent have looked at Kimi Code CLI (Kimi Code CLI — free Claude Code alternative, terminal-first agent), and one agency stack pairs Claude Code with Google Antigravity (AI tools worth paying for — Claude, Gemini, n8n, Neon and Railway…).

Several creators run two of these side by side. One builds a dashboard that drives both Claude Code and Codex subscriptions from one screen (Subscription-powered dashboards — custom UIs that drive Claude Code…), and a trending reverse-engineering repo works with either (Reverse Engineer Anything — trending repo that explains how any app…).

We counted every saved post in the DejaViewed corpus that names each tool, then sorted those posts by what the creator was doing. 129 entries name at least one of the three. Counts are computed from the live catalog each time this page is built.

Claude Code 126 entries

25 S · 46 A · 45 B · 10 C

Codex 18 entries

1 S · 7 A · 6 B · 4 C

Gemini CLI 2 entries

1 S · 1 A

The corpus barely covers Gemini CLI. The main entry is a 2025 post about the free tier that Google has since ended.

Method: a tool counts when it is named anywhere in an entry, including its caption and transcript. A theme counts when it appears in the entry's title, summary, tags, tools or takeaways. One entry can count toward several tools and themes. Tiers are our S/A/B/C quality ratings.

Which one to pick

All three now share the same building blocks: hooks, Agent Skills, plugins or extensions, MCP, subagents and a headless mode. The differences that matter are price of entry, license, and which ecosystem your team already lives in. We did not run our own benchmark, so this page does not rank code quality.

PickIf
Claude CodeYou already pay for Claude Pro or Max, you want the richest set of hooks, skills and plugins (see our best Claude Code skills and plugins), and you want a 1M-token context by default.
CodexYou want to start free on ChatGPT Free or $8 Go, you want an Apache-2.0 CLI you can read and fork, or you want a second model to audit what Claude writes.
Gemini CLIYour company already has a Gemini Code Assist Standard or Enterprise license or Vertex billing. Individuals on a personal Google account are pointed to Antigravity CLI instead.
Two of themYou plan with one model and audit with the other. Claude Code plus Codex is the pairing the corpus shows most, through OpenAI's Codex plugin for Claude Code.

What we could not verify

All comparisonsBest Claude Code skills and pluginsBest MCP serversSystem prompt examples

Sources

  1. Claude plans and pricing (Anthropic) accessed 2026-10-22
  2. Claude Code overview (Anthropic docs) accessed 2026-10-22
  3. Claude Code model configuration (Anthropic docs) accessed 2026-10-22
  4. anthropics/claude-code on GitHub: license, stars, releases accessed 2026-10-22
  5. Claude Code hooks reference accessed 2026-10-22
  6. Claude Code Agent Skills accessed 2026-10-22
  7. Claude Code plugins accessed 2026-10-22
  8. Claude Code MCP accessed 2026-10-22
  9. Claude Code subagents accessed 2026-10-22
  10. Claude Code agent view: background and parallel sessions accessed 2026-10-22
  11. Claude Code CLI reference (claude -p) accessed 2026-10-22
  12. Claude Code memory (CLAUDE.md) accessed 2026-10-22
  13. Use the Claude Agent SDK with your Claude plan (Anthropic support) accessed 2026-10-22
  14. API credits for Max and Team plans (Anthropic support) accessed 2026-10-22
  15. Codex pricing (OpenAI docs) accessed 2026-10-22
  16. Codex models (OpenAI docs) accessed 2026-10-22
  17. Codex IDE extension (OpenAI docs) accessed 2026-10-22
  18. openai/codex on GitHub: license, stars, releases accessed 2026-10-22
  19. Codex hooks (OpenAI docs) accessed 2026-10-22
  20. Codex skills and plugins (OpenAI docs) accessed 2026-10-22
  21. Codex MCP (OpenAI docs) accessed 2026-10-22
  22. Codex subagents (OpenAI docs) accessed 2026-10-22
  23. Codex git worktrees (OpenAI docs) accessed 2026-10-22
  24. Codex non-interactive mode (OpenAI docs) accessed 2026-10-22
  25. Codex AGENTS.md (OpenAI docs) accessed 2026-10-22
  26. An important update: transitioning Gemini CLI to Antigravity CLI (Google Developers Blog, 2026-05-19) accessed 2026-10-22
  27. google-gemini/gemini-cli on GitHub: README banner, license, stars, releases accessed 2026-10-22
  28. Gemini Code Assist quotas (Google Cloud docs) accessed 2026-10-22
  29. Gemini CLI quotas and pricing doc (repo; pre-June 2026 tiers) accessed 2026-10-22
  30. Gemini CLI model selection doc accessed 2026-10-22
  31. Gemini API models (Google) accessed 2026-10-22
  32. Antigravity pricing (Google) accessed 2026-10-22
  33. Gemini CLI hooks doc accessed 2026-10-22
  34. Gemini CLI skills doc accessed 2026-10-22
  35. Gemini CLI extensions doc accessed 2026-10-22
  36. Gemini CLI MCP servers doc accessed 2026-10-22
  37. Gemini CLI subagents doc accessed 2026-10-22
  38. Gemini CLI git worktrees doc (experimental) accessed 2026-10-22
  39. Gemini CLI headless mode doc accessed 2026-10-22
  40. Gemini CLI GEMINI.md doc accessed 2026-10-22
  41. Gemini CLI IDE integration doc accessed 2026-10-22