Short answer: Codex if you already pay OpenAI or want the current leaderboard lead, Claude Code if you want Claude models
Verified September 16, 2026. TL;DR: Codex is bundled into every ChatGPT plan, including the $0 one. Claude Code is included in every paid Claude plan and absent from the free one. Above that the ladders match: $20, then $100 or $200 for 5x or 20x usage. Price rarely decides it. Pick by the models you want and the subscription you already hold; Codex now holds the leaderboard lead, by 0.3 points.
| Codex | Claude Code | |
|---|---|---|
| Made by | OpenAI | Anthropic |
| Free tier | ChatGPT Free, $0 | None |
| Entry paid tier | Go $8, Plus $20 | Pro $20, or $17 billed annually |
| Top individual tier | Pro, $100 (5x Plus) or $200 (20x) | Max, $100 (5x Pro) or $200 (20x) |
| Models | GPT-6 Astra, GPT-5.6 Sol, Terra and Luna, plus GPT-5.5 until it retires Oct 14, 2026 | Sonnet 5, Opus 5, Fable 5.1, Haiku 4.5 |
| What it meters | Five-hour window; weekly limits may apply | Five-hour session plus a weekly limit |
| Runs in | Web, desktop app, CLI, IDE extension, iOS, cloud | Terminal, VS Code, JetBrains, desktop app, web, iOS and Android |
What is cheaper, Codex or Claude?
Codex, but only at the bottom of the ladder. ChatGPT Free includes it at $0, Go costs $8 and Plus $20. Claude Code starts at Pro, $20 a month or $17 billed annually, and Anthropic's Free plan does not include it. From $20 up the two match to the dollar: $100 and $200 for the 5x and 20x tiers, and team seats at $20 annual or $25 monthly on both sides.
Two meters that both count time
Both vendors meter five-hour windows, but only OpenAI prints numbers. On Plus, Codex allows 5 to 45 GPT-6 Astra messages per window, 10 to 100 on GPT-5.6 Sol, 25 to 200 on Terra and 250 to 2,000 on Luna; local and cloud work share the window, and weekly limits may apply. Anthropic publishes no counts: every plan resets on a rolling five-hour session, paid plans add a weekly limit, and chat, desktop and mobile share one pool with Claude Code.
Which is better, Claude Code Max or Codex Pro?
Same price, different fine print on the top model. Codex Pro is 5x or 20x Plus: 25 to 225 Astra messages and 50 to 500 Sol per window at $100, 100 to 900 Astra and 200 to 2,000 Sol at $200, plus GPT-5.3-Codex-Spark in research preview. Claude Max is 5x or 20x Pro per session, and it is the individual plan where Fable comes out of the included allowance, at up to half the weekly limit; on Pro, Fable bills to usage credits instead.
Can Codex do what Claude Code does?
Most of it. Both run in the terminal, an IDE and the cloud; both take MCP servers, skills, subagents and scheduled tasks; both sell credits to keep working past a limit. The gap is the model behind the agent. On Terminal-Bench 4.0, released August 28, Codex with GPT-6 Astra leads at 58.2%, Claude Code with Fable 5.1 is second at 57.9%, Opus 5 third at 51.8%, and GPT-5.6 Sol sixth at 37.3%.
How to choose, in one line each
Already paying for ChatGPT, run Codex for nothing extra. Want Claude models, buy Pro; want Fable inside the allowance, buy Max. Adding Copilot? Read the three-way breakdown. Cursor? Codex vs Cursor vs Claude Code. An open source agent? OpenCode vs Claude Code vs Codex. Google's free CLI? Gemini CLI vs Claude Code vs Codex. Not sure you should write the code at all? Run the vibe code or hire check.
Top comments (0)