Codex vs Claude Code: Price and Limits
Plans, prices, five-hour windows, weekly limits, per-model allowances and credits: the full map for choosing a tier between Codex and Claude Code.
Two coding agents dominate the conversation in August 2026: Codex, OpenAI's agent built into ChatGPT, and Claude Code, Anthropic's terminal tool. Both write, review and ship code inside your workflow, and both are paid through the same subscription as each company's chat. The practical question has moved on: how much each tier costs and where the limits start to bite. This guide lays out prices and limits as they appeared on the official pages in early August 2026, with one honest caveat: these figures move, so check the official page before deciding.
The entry point: twenty dollars on both sides
The first paid tier costs the same at both companies: twenty dollars a month.
On Anthropic's side, Claude Pro costs $20 per month billed monthly, or $17 per month when you pay for the full year up front ($200 in a single charge). That plan includes Claude Code and Claude Cowork alongside the chat, plus projects, Research and access to more models.
On OpenAI's side, ChatGPT Plus costs $20 per month, billed monthly. As of that date OpenAI offered annual plans only for Business accounts: Plus, Go and Pro were all month to month. Codex comes included in every ChatGPT plan starting from Free, and Plus is sized for a few focused coding sessions each week, with access to the Codex app for macOS and Windows and plugins that connect tools like Slack, GitHub and Figma.
The first difference shows up right here: the same entry price, different payment terms. Anthropic rewards an annual commitment with a discount; OpenAI kept the whole individual ladder on pure monthly billing.
The upper ladder: one hundred and two hundred, mirrored
Above the entry tier, both companies built the same ladder: a $100 level and a $200 level, with identical usage multipliers.
Claude Max starts at $100 per month and lets you choose between 5x or 20x more usage than Pro, always billed monthly. It adds higher output limits for every task, early access to advanced features and priority access at high traffic times.
ChatGPT Pro splits exactly the same way: the $100 tier unlocks 5x the usage of Plus, and the $200 tier unlocks 20x. Both Pro tiers share the same capabilities (Pro models, Codex, deep research, image creation, memory, file uploads); what separates them is usage allowance, and you can switch tiers anytime from your account settings.
The parallel is hard to miss: both companies concluded that heavy coding-agent users sort into two groups, the daily long-session crowd and the live-inside-the-tool crowd, and both priced those groups identically.
How limits bite in Claude Code
Here is the difference that matters day to day, more than price: the mechanics of the limit.
At Claude, every plan carries usage limits that reset on a rolling five-hour session window, and paid plans add weekly limits on top. Everything you do in Claude (web, desktop, mobile and Claude Code in the terminal) draws from the same pool. How much work fits in a window depends on the length and complexity of your conversations, the model you pick and the features you use, so the available message count varies instead of staying fixed.
When you reach a limit there are three paths: wait for the window to reset, move up a plan, or turn on usage credits, which on paid plans let you keep working at standard API rates. You can see where you stand anytime in Settings, under Usage.
One design detail with practical consequences: because the pool is shared, a heavy morning of coding in the terminal leaves less headroom for afternoon chats. Coding sessions consume a lot, since every turn drags project context along. If your pattern mixes chat and code through the day, that coupling weighs on which tier fits.
How limits bite in Codex
On OpenAI's side, Codex limits vary by plan and are framed around work sessions more than hourly windows. The official description of each tier says it plainly: Plus covers a few focused coding sessions per week; Pro is designed for longer, higher-intensity sessions each day.
Beyond the general limit, some models carry their own allowances. When you reach a model's allowance, that model becomes temporarily unavailable until the allowance resets, and ChatGPT displays the reset time when available; meanwhile you can keep going with another available model. The $100 and $200 Pro tiers share capabilities but assign different per-model amounts, and the platform's usage guardrails can temporarily restrict an account over automated patterns.
The expansion path exists here too: Codex credits let you keep working after hitting your plan limits, consumed based on the models and features you use. And OpenAI's API is billed independently of the ChatGPT subscription, as a separate product.
The free floor: a real difference
The sharpest contrast between the two sits at the free level. Codex is included across ChatGPT plans, including Free and Go: the free plan and Go carry limited trial access to Codex, enough to try the agent with the flagship model of the moment, GPT-5.6 at that time, with limited messages and uploads.
Claude also has a free chat plan, yet Claude Code is included in all paid plans and the official emphasis lands there: the terminal tool lives on the other side of the first charge. Evaluating Anthropic's agent on your own project means paying the first month; evaluating OpenAI's can start from the free plan.
For a serious evaluation in either case, the guide on testing a tool on your own project walks through the method step by step.
The surfaces each agent lives on
Price and limits tell half the story; the other half is where each agent actually runs. Codex spreads across several clients: the ChatGPT desktop app in Codex mode, the Codex CLI for the terminal, an IDE extension, and Codex on the web, all tied to the same ChatGPT account. Claude Code centers on the terminal, with VS Code and JetBrains integrations that bring the session next to the code, plus the plugins ecosystem that packages commands, subagents, hooks and MCP servers into installable units, which the plugins and extensions review details separately.
Both approaches end up in similar places: the agent sits where you already work, and the subscription follows you across surfaces. What changes is the center of gravity, terminal-first with editor companions at Anthropic, account-first with several equal clients at OpenAI.
Teams: seats and administration
For team work, Claude offers the Team plan: the standard seat costs $20 per person per month billed annually ($25 monthly), and the premium seat costs $100 ($125 monthly), with 5x the usage of standard seats and Claude Code and Cowork included. The plan covers teams of 2 to 150 people and adds central billing, single sign-on and admin controls over connectors.
OpenAI organizes its team offering around the Business and Enterprise plans, with per-seat billing and workspace controls: admins set Codex's starting model, reasoning level and plugin installation for each role, all from workspace settings. Seat prices appear on the official pricing page and are worth confirming there, since that part of the catalog adjusts frequently.
The API as a safety net
Both companies keep the API as a pay-per-use alternative, handy when work patterns are irregular or when the plan limit bites midweek.
At Anthropic, the model aimed at coding and agents carried introductory pricing at that time of $2 per million input tokens and $10 per million output tokens, valid through August 31, 2026, with standard rates of $3 and $15 after that date. Usage credits on paid plans are consumed precisely at those API rates.
At OpenAI, the API is contracted and paid separately from any ChatGPT subscription, with its own per-model price table.
Reading the whole picture
From above, the market converged: twenty dollars to start seriously, one hundred and two hundred for intensive work, the coding agent bundled with the chat subscription, and the API as a pay-per-use path. The remaining differences are matters of mechanics and nuance.
Free access to Codex makes it easy to try before paying; Claude Code asks for the first tier from the start. Claude Pro's annual discount rewards consistency; OpenAI's individual ladder stayed purely monthly. Claude's limits organize around five-hour windows with a weekly cap and a pool shared between chat and terminal; Codex limits are expressed as weekly sessions per tier plus per-model allowances.
The practical advice in early August 2026 was to start at the twenty-dollar tier in the company whose ecosystem you already use, measure your consumption for two or three weeks (both platforms show current usage in settings) and move up only when the limit bites two weeks in a row. If the extension ecosystem is the deciding factor, the review of Claude Code plugins and extensions details that ground, and the guide on what skills are covers the piece both agents use to package procedures.
Two final practical notes. First, the figures quoted here come from the official pages as they stood between late July and the first days of August 2026, and both companies adjust their catalogs regularly, so treat the pages linked below as the living reference. Second, limits are easier to feel than to read about: a week of real work on the entry tier teaches more about which multiplier you need than any comparison table, and both platforms show the current consumption clearly enough to make that judgment with data in hand.
Sources
Was this article helpful?