Claude Code or Codex? How Their Prices and Usage Limits Compare

Claude Code and Codex now cost the same at the first paid tier. Both come with a $20-a-month subscription, both run in the terminal, in an editor and in the cloud, and both meter your work with a five-hour window and a weekly cap that neither company puts a number on. The differences show up elsewhere: who gets in for free, what the top tiers remove, how much each company tells you about your allowance, and what happens when it runs out.

Everything here comes from Anthropic's and OpenAI's pricing pages, help centers and product documentation and from the Codex changelog, checked on October 18, 2026, Beijing time. We haven't benchmarked the two tools against each other. Where a company describes its own models, we quote it.

Same $20 first paid tier, different ladders

Anthropic doesn't include Claude Code on its Free plan. The first rung is Pro at $20 a month, or $17 a month if you pay $200 for the year. Above it sit Max 5x at $100 and Max 20x at $200, both billed monthly.

OpenAI's ladder is longer at both ends. Codex is included on ChatGPT Free and on Go, which costs $8 a month, but only with GPT-6 Luna at Standard speed in the desktop app, subject to rollout, and without Codex Cloud. Plus, at $20, adds Codex on the web, in the CLI, in the IDE extension and on iOS, cloud integrations such as automatic code review and Slack, and GPT-6.1 Sol. Pro comes in three monthly tiers at $100, $200 and $500. OpenAI says Pro $200 includes more usage than Pro $100 and Pro $500 the most, without giving amounts, and only Pro $500 gets Ultrafast mode. If you're weighing Pro $200, OpenAI says new subscriptions now get a lower allowance than before; eligible existing subscribers keep the old one through October 29, 2026.

For teams, the two structures look alike. Claude Team charges $20 per seat per month billed annually, or $25 monthly, for a Standard seat, which Anthropic says gets more usage than Pro, and $100 or $125 for a Premium seat with five times a Standard seat's usage. ChatGPT Business charges $20 per user per month billed annually, or $25 monthly, for a Standard seat, and $100 billed annually or $125 monthly for a Premium seat, which OpenAI says gets five times Standard's usage with no five-hour limit.

What the allowance buys

Neither company publishes a weekly figure, so the useful comparison is what each one does disclose. OpenAI gives estimates of local messages per five-hour window for Plus and Business Standard, broken out by model: 5 to 45 with GPT-6 Astra, 15 to 160 with GPT-6.1 Sol, 15 to 150 with GPT-6 Sol and 350 to 3,000 with GPT-6 Luna. It calls these estimates rather than fixed limits, says cloud tasks may use more, and says weekly limits may also apply. Pro plans currently have no five-hour limit.

Anthropic publishes only multiples. Its pricing page says Pro gives at least five times the Free plan's usage per five-hour session, and Max gives five or 20 times Pro's. Paid plans also have weekly limits, and Max adds a separate weekly limit for Anthropic's Fable models, which can take up to 50% of the weekly allowance. On Pro, Fable runs only on usage credits.

So the published figures can't tell you which $20 plan buys more coding time; the units don't line up. What both companies make clear is that model choice moves consumption more than anything else. OpenAI's credit rate card charges 250 credits per million input tokens and 1,250 per million output tokens for GPT-6 Astra, against 50 and 250 for GPT-6.1 Sol and 2.5 and 12.5 for GPT-6 Luna, and Fast mode uses included limits at 2.5 times the Standard rate. On the Claude side, Claude Code defaults to Opus 5.5 at medium effort on Pro, Max, Team and Enterprise.

Both allowances are shared with the rest of the subscription, which matters if you also chat a lot. Anthropic says Claude on the web, desktop and mobile and Claude Code all draw from the same pool. OpenAI says ChatGPT Work and Codex share pricing, credits and usage limits, and on Plus and Pro the pool also covers ChatGPT for Excel.

When you run out

Both tools try not to stop halfway through a task. OpenAI says Codex can keep working on a turn that is already in progress when you reach a limit, subject to fair-use limits. Anthropic is gradually rolling out a small, capped wrap-up allowance that lets Claude Code finish the current step when you reach your five-hour limit: up to once per weekly period on Pro, and each time on Max.

After that the options diverge. On Claude Pro and Max you can turn on usage credits, billed at standard API rates up to a monthly spend limit you set. Anthropic also hands out free limit resets from time to time, and its help center doesn't describe any way to buy one. OpenAI gives Plus and Pro users more levers. There are credits, whose purchase price OpenAI says depends on your plan or agreement. There are free banked resets that come with some promotions and referrals. And some accounts can buy an instant reset, which restores both windows straight away but, in OpenAI's words, pulls your normal weekly allowance forward rather than adding a separate entitlement; your next weekly reset moves to seven days after your next request.

Where each one runs

The two tools cover similar ground. Claude Code runs in the terminal, in VS Code, Cursor and JetBrains IDEs, in a desktop app and in the browser, including through the Claude mobile apps. Its Routines run in Anthropic's cloud on a schedule or in response to GitHub events, even when your computer is off, and Claude Code also works with GitHub Actions, GitLab CI/CD, GitHub code review and Slack. Codex runs in the ChatGPT desktop app, the CLI, an IDE extension that OpenAI says works with most VS Code forks, on the web and on iOS. Codex Cloud runs tasks on OpenAI-managed computers that keep going while yours is asleep, and Codex connects to GitHub for pull request reviews, as well as to Slack and Linear.

Two other differences are practical. Anthropic documents running Claude Code's terminal, VS Code and JetBrains versions through Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry, which suits companies that already buy models through a cloud provider; for Codex, the changelog we checked mentions Amazon Bedrock catalogs. And if you use both tools, the instructions can be shared: Codex reads AGENTS.md, and Anthropic says Claude Code can read an existing AGENTS.md in place of its own CLAUDE.md.

Paying by the token instead

Either tool can run without a subscription. OpenAI says you can use Codex in the CLI, SDK or IDE extension with an API key, billed at API rates, with access to whatever models the key can reach but no cloud features such as GitHub code review or Slack. Anthropic supports API keys from the Claude Console, Amazon Bedrock, Google Cloud and Microsoft Foundry, billed per token. Its cost guide says enterprise deployments average around $13 per developer per active day and $150 to $250 per developer per month, with 90% of users staying under $30 per active day. Anthropic also warns that if an ANTHROPIC_API_KEY environment variable is set, Claude Code uses that key instead of your subscription and bills you at API rates.

Which one to pick

If you want to try a coding agent without paying, Codex is the only one of the two with a way in, though Free and Go are limited to GPT-6 Luna in the desktop app. At $20, the deciding question is often which subscription you already rely on, because coding comes out of the same pool as your chats. Codex Plus comes with per-model ranges you can plan against; Claude Pro discloses less, but puts Anthropic's default Opus 5.5 model in Claude Code.

Heavy individual users face a sharper choice. If the five-hour window is what keeps stopping you, Codex Pro removes it for now, while Claude Max keeps the window and multiplies what fits in it. Codex Pro also goes up to $500 a month, with Ultrafast mode, while Claude's individual plans top out at $200 and Max is where Fable usage is included without credits. Teams will find the seat structures close to interchangeable.

On model quality, we can only pass on what the companies say. OpenAI's changelog says "GPT-6.1 Sol offers near-Astra performance for complex work at a lower cost than Astra." Anthropic's pricing page labels Opus 5.5 "Daily driver for agentic coding and enterprise work" and Fable 5.1 "Next generation intelligence for long-running agents." Both companies change these plans often, and our separate usage-limit explainers for each tool track the details.