Codex vs Claude Code pricing: what your subscription actually buys

Codex through ChatGPT Plus and Claude Code through Claude Pro both start at USD20 a month for individual CLI coding use. The better buy depends on which agent completes your work within its allowance, not which provider advertises the larger multiplier.

This comparison checks prices and billing boundaries on 29 October 2026. It is a buying checklist, not a benchmark or a promise of tokens per dollar. US monthly prices below are reference prices; check currency, tax and purchase platform at checkout.

How much does each subscription cost?

OpenAI’s pricing documentation and Pro tiers help list the ChatGPT options. Anthropic lists Claude Pro and Max separately.

Individual monthly option	USD reference price	Buying consideration
ChatGPT Plus	$20	Starting point for Codex CLI or IDE use
ChatGPT Pro 100 / 200 / 500	$100 / $200 / $500	Higher tiers; Pro 500 adds Ultrafast
Claude Pro	$20	Starting point for Claude Code
Claude Max 5x / 20x	$100 / $200	Larger Claude allowance; still limited

OpenAI also lists Free and Go, but its current pricing page describes those Codex options as Luna in the desktop app, subject to rollout. They are not equivalent substitutes for the CLI workflow compared here.

If you already pay for one provider, test that subscription before buying the other. A second subscription only helps if it solves a specific constraint: missing model access, a required integration, or work that the first agent repeatedly fails to finish.

Why can equal monthly prices buy different working time?

OpenAI’s usage guidance says Pro currently has no five-hour limit; that does not mean unlimited weekly work. Its Plus local-message estimates are model-dependent, not guaranteed quotas. Claude’s plan help describes shared Claude and Claude Code limits, including five-hour and weekly windows.

Do not turn either provider’s plan multiplier into a cross-provider token budget. Instead, define a unit of useful work: a bug fixed with a regression test, a migration reviewed, or a feature accepted after manual review. Count retries and abandoned runs as part of the cost.

A simple comparison worksheet is:

• Task and acceptance test, written before running either agent.
• Same repository starting point and relevant context.
• Model, effort and speed setting used by each agent.
• Whether the result passed tests and needed manual repair.
• Remaining allowance before and after, plus any additional paid usage.

Repeat across different task types. One easy success or one frustrating failure should not determine a recurring subscription purchase. This is our suggested evaluation method, not a test we have run.

Does signing in mean you are using the subscription?

OpenAI authentication docs distinguish ChatGPT sign-in from API-key access. Anthropic’s Claude Code account guidance warns that an ANTHROPIC_API_KEY environment variable takes precedence and can produce API charges.

Before a trial, check the active account and billing route. Do not paste keys, tokens or credential files into a support thread. Record only non-secret account and usage information.

Claude Max also has claimable monthly API credits, separate from interactive Claude Code allowance. Treat an API-backed agent as its own budget, rather than assuming those credits replenish a depleted Claude Code session.

For token-cost arithmetic, use our earlier Claude subscription versus API guide, but recheck its dated rates before making a purchase. This page compares subscription choices, not API session prices.

Which apps let you compare both agents without changing editors?

T3 Code is one option: its README describes controlling locally authenticated Claude Code and Codex agents. Conductor also supports both; its FAQ says it normally uses authentication already stored on the machine, including an existing API-key route.

Neither control surface turns a ChatGPT plan into a Claude plan. Choose the intended agent and verify its authentication separately. Anthropic’s legal documentation distinguishes a user signing into unmodified Claude Code from developers routing their users’ consumer credentials through their own apps.

Browse the Claude-compatible apps and ChatGPT-compatible apps by the plan you already own. For the native account boundary, see using a ChatGPT subscription in Claude Code.

What should you buy after the comparison?

Keep the least expensive plan that reliably completes your important work. Upgrade when your recorded interruptions are expensive enough to justify it, and when the higher tier addresses that particular limit. Keep both providers only if each has a recurring job the other does not cover well enough.

For an existing Pro 200 account, check OpenAI’s current tier notice: eligible subscribers keep the previous allowance through 29 October, then move to a lower allowance at the same price. Do not use a pre-transition trial as a permanent capacity promise. The notice does not establish an exact Singapore transition hour.

A useful decision is “this plan completes my weekly review workload without extra spending,” not “this plan has the biggest number next to its name.”