AI Tools

Claude Code vs Codex: which AI coding agent should you pay for?

Published 2026-08-23 · Researched and price-checked on official sources

Claude Code vs Codex: both coding agents reach a $20/month tier, but they meter usage differently. Claude Code caps weekly usage, boosted 50% through August 31, 2026; Codex estimates messages per five-hour window by model. Codex reads AGENTS.md natively; Claude Code imports it through CLAUDE.md. Prices verified on official pages August 23, 2026.

Key facts

What matters

Claude Code has no standalone price: it ships inside Claude Pro ($20/mo monthly, $17/mo annual), Max (from $100/mo), and Team plans. Codex starts free, with Go at $8/mo, Plus at $20/mo, and Pro from $100/mo.
Metering is the real differentiator: Claude Code draws from a weekly limit plus a 5-hour cap (weekly limits are 50% higher through Aug 31, 2026); Codex allows 10-100 GPT-5.6 Sol messages per five-hour window on Plus, up to 5,000-40,000 Luna messages on Pro 20x.
AGENTS.md compatibility just changed: Codex reads AGENTS.md before doing any work, while Claude Code reads CLAUDE.md and needs an @AGENTS.md import or symlink. The feature request (issue #6235, 375 comments) closed as completed on August 17, 2026.
Both sell the same escape hatch from caps: an API key with per-token billing, or ChatGPT credits on the Codex side.

What are Claude Code and Codex?

Claude Code is Anthropic's coding agent that runs in the terminal CLI, IDE extensions, the desktop app, and the web. It is not sold separately: it comes with Claude Pro, Max, and Team subscriptions, per claude.com/pricing.

OpenAI Codex is OpenAI's coding agent, available in the ChatGPT web app, the Codex CLI, the IDE extension, and iOS. Per the official pricing docs, it is included in ChatGPT Free, Go, Plus, Pro, Business, Edu, and Enterprise plans, and it shares its usage budget with ChatGPT Work.

Both agents do the same headline job: read your repository, run commands, edit files, and complete multi-step coding tasks from a natural-language prompt. The differences that decide the purchase are pricing structure, usage metering, and which project-instructions file each one honors.

Claude Code vs Codex pricing, verified

Claude Code prices were verified on claude.com/pricing on August 23, 2026. The cheapest way in is Claude Pro at $17/month billed annually ($200 up front) or $20/month billed monthly, which includes Claude Code, Claude Cowork, and chat usage. Max starts at $100/month and offers 5x or 20x more usage than Pro. Team plans start from $20 per seat per month billed annually.

Codex prices were verified on the official ChatGPT pricing docs the same day. The Free tier ($0) covers quick coding tasks, Go costs $8/month for lightweight tasks, Plus costs $20/month for "a few focused coding sessions each week," and Pro starts at $100/month with 5x or 20x higher rate limits than Plus. Plus unlocks the GPT-5.6 model family (Sol, Terra, Luna), cloud integrations like automatic code review and Slack, and the option to buy extra ChatGPT credits when limits run out.

The honest comparison: Claude Code's entry floor is $17-$20/month with no free tier, while Codex has a genuinely usable free tier and an $8 Go plan. At the heavy-use end both vendors charge from $100/month, and both are capped rather than unlimited.

Usage limits: weekly cap vs five-hour windows

Claude Code on a Claude plan draws from a weekly usage limit plus a 5-hour usage limit, per Anthropic's support documentation. Anthropic raised the weekly limits by 50% for Pro, Max, and Team plans from May 13 through August 31, 2026, applied automatically; Free plans and consumption-based Enterprise seats get no boost. Anthropic does not publish a fixed weekly hour figure, and the /usage command inside Claude Code shows your current limit.

Codex measures usage differently: the official docs estimate local messages per five-hour window, and the allowance scales with model choice. On Plus, GPT-5.6 Sol allows 10-100 messages, GPT-5.6 Terra 25-200, and GPT-5.6 Luna 250-2,000. Pro 5x multiplies those ranges by roughly 5 (Sol 50-500, Luna 1,250-10,000), and Pro 20x by roughly 20 (Sol 200-2,000, Luna 5,000-40,000). Cloud chats run on Sol and consume more of the budget, and additional weekly limits may apply.

The catch on both sides: hitting the cap mid-task. Codex lets you buy ChatGPT credits to continue without upgrading, and any user can run extra local chats through an API key at standard API rates. Claude Code users can sign in with an API key instead of a plan, which removes the weekly cap entirely and bills per token.

AGENTS.md: the compatibility gap that just closed

AGENTS.md is the project-instructions file that OpenAI's ecosystem standardized on. OpenAI's Codex documentation states plainly that "Codex reads AGENTS.md files before doing any work," layering global guidance with project-specific overrides.

Anthropic's docs, by contrast, say Claude Code reads CLAUDE.md, not AGENTS.md. The supported bridge: create a CLAUDE.md that imports the existing AGENTS.md with an @AGENTS.md line, or symlink the file (ln -s AGENTS.md CLAUDE.md); Windows users should use the import syntax. Claude Code v2.1.213 or later also ships an /import command that copies AGENTS.md into CLAUDE.md and carries over MCP servers, commands, subagents, and skills.

The timing is the story. The request to support AGENTS.md in Claude Code (issue #6235 on GitHub) gathered 375 comments and was closed as completed on August 17, 2026, days before this article. A team that keeps one AGENTS.md for its repo can now feed both agents from that single file, with Claude Code needing only the one-line import.

Which should you choose?

Choose Claude Code if you already pay for Claude Pro or Max: the agent is included, and teams on Claude Team get per-seat access with no separate license. It fits developers who live in a terminal or IDE, want subagents and skills, and can schedule around a weekly cap that is currently 50% more generous than usual.

Choose Codex if you want a free tier to start, an $8 Go plan for light work, native AGENTS.md handling, or the Luna path to high-volume tasks: the message allowances on Luna dwarf the other tiers. Plus buyers also get cloud-based automatic code review and Slack integration that Claude Code does not bundle.

Neither is the unlimited option. Serious daily users of either agent end up at the $100/month tier, and the true ceiling-remover on both platforms is an API key with pay-per-token billing.

At a glance

	Claude Code	OpenAI Codex
Entry price	$20/mo monthly or $17/mo annual (Claude Pro)	Free tier; Go $8/mo; Plus $20/mo
Heavy-use tier	Max from $100/mo (5x or 20x usage)	Pro from $100/mo (5x or 20x rate limits)
Usage metering	Weekly limit + 5-hour limit; weekly +50% through Aug 31, 2026	Messages per 5-hour window by model; credits for overage
Project instructions	CLAUDE.md; AGENTS.md via @import or symlink	AGENTS.md read natively
Models	Claude Mythos family (Opus, Sonnet, Haiku)	GPT-5.6 Sol, Terra, Luna; GPT-5.3-Codex-Spark (Pro preview)
Overage path	API key, per-token billing	ChatGPT credits or API key

Claude Code vs Codex, verified on official pages, August 23, 2026

FAQ

Is Codex free?

Yes, there is a free tier for quick coding tasks. Paid Codex tiers are Go at $8/mo, Plus at $20/mo, and Pro from $100/mo, per the official ChatGPT pricing docs verified August 23, 2026.

Does Claude Code support AGENTS.md?

Not natively. Claude Code reads CLAUDE.md, so you add an @AGENTS.md line inside CLAUDE.md or symlink the file, or run /import (Claude Code v2.1.213+) to copy AGENTS.md into CLAUDE.md. Feature request #6235 closed as completed on August 17, 2026.

Which is cheaper for daily coding, Claude Code or Codex?

At the entry tier they are close: Claude Pro is $20/mo monthly or $17/mo annual, Codex Plus is $20/mo. The real difference is metering: Claude Code runs on a weekly cap (currently +50% through Aug 31, 2026), while Codex allows roughly 10-100 GPT-5.6 Sol messages per five-hour window on Plus. Heavy users on both usually move to the $100/mo tier or an API key.