
Codex vs. Claude Code Pricing and Usage Limits in 2026
By Nathan Cole
MyClaw Editorial
MyClaw
Run Best-in-Class AI Agents Now
Run OpenClaw or Hermes Agent on managed MyClaw hosting, so your AI agent stays online, updated, and ready for real work.
AI Takeaway
- Which coding agent costs less to try? ChatGPT Free includes limited Codex access. Claude Code subscriptions start at $20 a month with Claude Pro, or $200 if you pay annually. Claude Console offers pay-as-you-go access.
- How far do the two $20 plans go? ChatGPT Plus includes Codex with a five-hour usage window shared with ChatGPT Work. Claude Pro includes Claude Code with five-hour and weekly limits shared with Claude chat. The work you can finish depends on the model, the size of the task, and what else draws from your plan.
- What happens when you hit a limit? Wait for a reset or buy extra usage on an eligible plan. OpenAI uses its ChatGPT credit rates. Anthropic charges standard API rates when you enable extra usage on your Claude subscription.
- How should you budget for automated coding? Use a trial run to estimate the API charges, then multiply by the number of runs you expect each month and add any hosting fees.
Codex and Claude Code Plans Compared
Free Access and the $20 Plans
Here are the U.S. web prices for individual subscriptions as of November 16, 2026.
| Monthly price | Codex with ChatGPT | Claude Code with Claude |
|---|---|---|
| $0 | Free, limited Codex access | Claude Free excludes Claude Code |
| $8 | Go, limited Codex access | — |
| $20 | Plus | Pro, or $200 billed annually |
| $100 | Pro 100 | Max |
| $200 | Pro 200 | Ultra |
| $500 | Pro 500 | — |
ChatGPT Free and Go offer limited access to GPT-6 Luna at standard speed in the desktop app, subject to rollout. Plus adds Codex access on the web, in the terminal, and in an IDE, along with cloud features such as automatic code review. Claude Pro includes Claude Code in the terminal and other supported clients. Paying $200 up front for Claude Pro saves $40 over a year of monthly payments.
If you’re choosing between Codex and Claude Code for daily development, check where you want the agent to work and how you’ll review changes.
What You Get at $100 and Up
Claude Max gives you five times Claude Pro’s usage in both the five-hour and weekly windows. At $200, Claude Ultra replaces Max 20x. It provides ten times Pro’s weekly usage and removes the five-hour limit for verified accounts.
Each step up from ChatGPT Pro 100 to Pro 200 to Pro 500 increases the shared Codex and ChatGPT Work allowance. Pro 500 also adds Ultrafast for supported models. When you use Ultrafast, the included allowance is consumed at eight times the rate of the same model at standard speed.
What the Usage Limits Mean for Daily Coding

Five-Hour Windows, Weekly Caps, and Shared Usage
For ChatGPT Plus, OpenAI estimates roughly 15–160 local Codex messages per five-hour window with GPT-6.1 Sol. Where you land in that range depends on your tasks. Codex cloud tasks and ChatGPT Work share the allowance. ChatGPT Pro removes the five-hour limit. Weekly limits can still apply on both Plus and Pro.
Claude Pro shares five-hour and weekly limits across Claude chat and Claude Code. If you spend the morning analyzing files in Claude and then switch to your terminal, you’re drawing from the same allowance. Anthropic measures its 5× and 10× allowances against Claude Pro. To compare them with Codex, track how much of your own work fits before each plan resets.
A one-file fix may take a short diff and one test run. A repository-wide refactor can require reading dozens of files, running tests across several packages, and keeping track of earlier decisions. Both can start with a one-line prompt and consume very different amounts of your allowance.
What Happens When You Run Out
Check the Codex usage dashboard or Claude’s Settings → Usage for your next reset. If you still have allowance, a lighter model can stretch it for routine fixes. Once you hit the cap, you can wait for the reset or pay for more usage.
ChatGPT Plus and Pro users can buy ChatGPT credits. Codex draws from that balance at rates that depend on the model and settings. Claude Pro, Max, and Ultra users can enable usage credits billed at standard API rates and set a monthly spending cap.
API Pricing for CI and Scheduled Coding Tasks

Which Account Pays for an Automated Run?
If you run scheduled coding jobs in OpenClaw, check how the coding agent authenticates before estimating its cost. Codex with a ChatGPT sign-in uses your plan allowance first, then any purchased credits available for eligible usage. If the Codex CLI or SDK authenticates with an API key, OpenAI meters the run at API rates in the organization that owns the key.
Eligible ChatGPT Pro subscribers can claim monthly API credits of $100, $200, or $500, matching their tier. Claude Max and Ultra subscribers can claim $100 and $200, respectively. These credits go to a linked developer organization and cover eligible API runs. Your purchased subscription credits cover extra interactive usage.
With Claude, a standalone claude -p run using a key from your linked Console organization can draw from the monthly API grant. The same command with a Claude subscription sign-in draws from your plan’s limits. Anthropic classifies runs started by its Claude Code GitHub Action, an IDE extension, or the desktop app as Claude Code usage, outside the monthly API grant.
What 40 Automated Reviews Could Cost
At standard API rates, GPT-6.1 Sol and Claude Sonnet 5.5 each cost $2 per million uncached input tokens and $10 per million output tokens. If a review uses 100,000 uncached input tokens and 10,000 output tokens, the model charge is $0.20 for input and $0.10 for output. That’s $0.30 per run, or $12 for 40 runs with the same usage, before credits, tool charges, or hosting.
For your own estimate, take the token totals from a few real reviews. Agents may load different files, cache repeated context, or retry a failed run, so equal token prices can still lead to different bills. Include OpenClaw hosting costs if you use a managed instance, then divide the total by the number of reviews you could use.
Which Coding Plan Is Worth Paying For?
For occasional fixes, try the limited Codex access on ChatGPT Free or pay by API usage if you need a specific model or client. If you write code most days, start with a week on the $20 plan you already have. Track changes you accepted, sessions interrupted by a cap, and credits you bought. If you try the other plan, use the same repository and a similar mix of fixes, tests, and reviews.
Moving from $20 to $100 costs another $80 a month. Weigh that against what you spend on extra usage and how often a reset delays your work. A daily pull request digest is a small code automation workflow you can price before adding full code reviews or CI debugging.
Run a Scheduled Pull Request Digest With MyClaw

MyClaw’s managed OpenClaw hosting keeps a private agent online and handles updates and backups, so your weekday GitHub digest can run while your laptop is off. Include the hosting fee and any charges from the models or coding agents the digest uses.
- Launch a private agent. Set up an OpenClaw instance in MyClaw and ask it to summarize pull requests opened since the previous run. For each PR, include a link, a short summary of the change, and a specific test gap if the diff shows one.
- Connect one repo and choose how to review it. Connect one GitHub repository with read-only access and choose the model and billing account for the summaries. The Coding Agent skill can delegate diff reviews to Codex or Claude Code once you install and authenticate its CLI in the hosted environment.
- Test, schedule, and check the cost. Run the digest once and verify its notes against the diffs. Then set a weekday schedule and send it to a team channel. Skip the review step when there are no new pull requests. After the first week, check the cost per digest and ask the team which findings saved them time.
Conclusion
Before your next renewal, compare a normal coding week with your busiest one. If you only paid for extra usage during a release, estimate what those occasional top-ups would cost over a few months before committing to a higher tier. For scheduled reviews, set a monthly budget that includes hosting. Add a second repository once the first digest consistently surfaces findings your team acts on.
Skip the Setup, Run Best-in-Class AI Agents Now
Launch a managed OpenClaw or Hermes Agent workspace in minutes, with always-on hosting, updates, and support handled by MyClaw.