Head to head
Codex vs Claude Code (2026): Claude Code vs Codex Pricing, Limits and Which Is Better
Two agents bundled into two chat subscriptions, priced identically at every rung. Figures read from OpenAI's and Anthropic's own pricing and documentation pages in August 2026.
No card to start · Review-first: the agent never merges
Pick a task
Plan
- planning
Files changed
Test run
Pull request
You review and merge. Agentcode never merges on its own.
The short answer
Codex and Claude Code cost exactly the same at every rung: $20 a month to start, $100 for the 5x tier, $200 for the 20x tier, and $20 per seat a month billed annually for a team. Both run a rolling five hour usage window with weekly limits on top, and on both the coding agent shares that allowance with the vendor's chat product. So price cannot decide this. The real difference is transparency. OpenAI publishes how many messages each model gives you per five hour window, 10 to 100 on GPT-5.6 Sol on Plus and 250 to 2,000 on Luna, plus a credit rate card per million tokens, so you can forecast spend before you buy. Anthropic publishes a multiplier instead, and answers the question from the other end with observed cost: around $13 per developer per active day and $150 to $250 per developer per month. Codex is also the only one of the two with a free tier.
What is the difference between Codex and Claude Code?
Structurally they are the same product idea from two labs: an agentic coding tool that reads your codebase, edits files and runs commands, sold as part of a chat subscription rather than on its own. The differences are in the details, and the details are where the money goes.
Swipe to see more
| OpenAI Codex | Claude Code | |
|---|---|---|
| What it is | OpenAI's coding agent, bundled into a ChatGPT subscription rather than sold on its own. It runs locally and in the cloud. | Anthropic's coding agent, bundled into a Claude subscription. Also runs locally and in the cloud. |
| Where you run it | Web, CLI, IDE extension and iOS, plus cloud integrations like automatic code review and Slack. | Terminal CLI, VS Code, JetBrains, a desktop app, the browser at claude.ai/code, and iOS and Android. |
| Cheapest way in | Free. Codex is included on the $0 ChatGPT plan, with an $8 Go tier above it. | $17 a month on annual Pro, or $20 billed monthly. There is no free Claude Code tier. |
| Models | The GPT-5.6 family: Sol for the hardest work, Terra as the everyday workhorse, Luna for fast high volume tasks. Older GPT-5.5 and 5.4 models remain selectable. | Anthropic's own Claude models only. There is no third party model routing. |
| How usage is counted | Credits, priced per million input, cached and output tokens, on a published rate card per model. | Usage against your plan, expressed as a multiple of the Pro allowance rather than a token rate card. |
| Limit structure | A rolling five hour window, with weekly limits on top. Local messages and cloud chats share the same window. | A rolling five hour window, with weekly limits on top. Shared with Claude chat and Cowork. |
| What else eats the same quota | ChatGPT Work, ChatGPT for Excel on Plus and Pro, Voice initiated tasks, and image generation, which drains limits three to five times faster. | Claude chat and Claude Cowork draw the same allowance. |
| When you hit the ceiling | Plus and Pro users can buy additional credits without upgrading the plan. You can also fall back to an API key at standard API rates. | Add usage credits to the subscription, move up a tier, or run against the Anthropic API instead. |
Codex vs Claude Code pricing: the ladders are identical
This is the finding that should change how you approach the decision. Lay the two lineups side by side and they match at every rung that exists on both: $20 to start, $100 for 5x, $200 for 20x, $20 a seat for a team billed annually. Both vendors even use the same words for the step up, offering "5x or 20x" more usage than the entry plan. If you came here to find the cheaper one, the honest answer is that there isn't one.
Swipe to see more
| Rung | OpenAI Codex | Claude Code | What it means |
|---|---|---|---|
| Free | Free, $0. Codex included. | Free, $0. No Claude Code. | The only rung where the two differ structurally. Codex is the only one of the pair you can run without paying. |
| Light | Go, $8 a month. | No equivalent. | OpenAI sells a rung below the $20 entry point. Anthropic does not. |
| Entry | Plus, $20 a month. | Pro, $20 a month, or $17 billed annually. | Identical sticker price. Anthropic discounts for annual billing, OpenAI does not at this tier. |
| 5x | Pro 5x, $100 a month. | Max 5x, $100 a month. | Same price, same multiplier, and both vendors use the literal phrase "5x or 20x". |
| 20x | Pro 20x, $200 a month. | Max 20x, $200 a month. | Same again. At the top of the individual ladder the two are indistinguishable on price. |
| Team seat | Business, $20 per user annually, $25 monthly. | Team, $20 per seat annually, $25 monthly. | Exactly the same, to the dollar, on both billing cycles. |
| Heavy team seat | No separate rung. Buy workspace credits. | Team premium seat, $100 annually, $125 monthly, for 5x a standard seat. | Anthropic sells a heavy seat as a product. OpenAI handles the same need with credits. |
| Bring your own key | Sign in with an API key and pay standard API rates. No cloud features. | Enterprise is $20 a seat plus usage at API rates, or use the Anthropic API directly. | Both let you bypass the subscription meter entirely and pay per token instead. |
The one asymmetry worth acting on sits at the bottom of the table rather than the top. Codex is included on the free ChatGPT plan and has an $8 Go tier underneath Plus. Claude Code starts at Pro and has no free tier at all. For an individual picking something to learn on, that is most of the decision. For a team running a paid evaluation, it changes nothing. See how the rest of the category structures the same choice in our flat versus credit metered pricing comparison.
Codex vs Claude Code usage limits: what each vendor will tell you
Both products run the same limit architecture: a rolling five hour window, with weekly limits on top. On Codex the five hour window is shared between local messages and cloud chats. On Claude Code it is shared with Claude chat and Cowork. Where they part company is how much they will tell you about it in advance.
OpenAI publishes a message count per model per five hour window. Anthropic publishes a multiplier. Here is the OpenAI table, reproduced from its documentation, for local messages. Note how wide each range is: the low end and the high end differ by a factor of ten, which is OpenAI telling you plainly that task shape matters more than plan choice.
Swipe to see more
| Model | Plus, per 5h | Pro 5x, per 5h | What it is for |
|---|---|---|---|
| GPT-5.6 Sol | 10 to 100 | 50 to 500 | Complex reasoning, ambiguous problems, advanced coding. |
| GPT-5.6 Terra | 25 to 200 | 125 to 1,000 | The everyday workhorse for production tasks and coding. |
| GPT-5.6 Luna | 250 to 2,000 | 1,250 to 10,000 | Fast, high volume work and focused coding tasks. |
| GPT-5.5 | 15 to 80 | 75 to 400 | Previous generation, still selectable. |
| GPT-5.4 | 20 to 100 | 100 to 500 | Cheaper per token than 5.5 at half the credit rate. |
| GPT-5.4 mini | 60 to 350 | 300 to 1,750 | The cheapest model on the rate card. |
Underneath that sits a credit rate card, priced per million tokens, which is what lets you cost a workload rather than guess at it. The spread is the interesting part: Luna costs five credits per million input tokens where Sol costs 125, so routing a simple task to the cheap model is a twenty five fold saving on input. Anthropic has no published equivalent, which is the single clearest practical difference between buying these two products.
Swipe to see more
| Credits per 1M tokens | Input | Cached input | Output |
|---|---|---|---|
| GPT-5.6 Sol | 125 | 12.5 | 750 |
| GPT-5.6 Terra | 50 | 5 | 300 |
| GPT-5.6 Luna | 5 | 0.5 | 30 |
| GPT-5.5 | 125 | 12.50 | 750 |
| GPT-5.4 | 62.50 | 6.25 | 375 |
| GPT-5.4 mini | 18.75 | 1.875 | 113 |
OpenAI adds that GPT-5.6 usage averages 5 to 40 credits per message, and that fast mode consumes credits at a higher rate. For a fuller breakdown of the Anthropic side, see our Claude Code pricing guide.
The six things that actually decide it
Benchmark posts will tell you which model scored higher last month. That changes with every release. These six do not.
The two price ladders are identical, so price is not the deciding factor
Put the plans side by side and something odd falls out: OpenAI and Anthropic have converged on exactly the same ladder for agentic coding. Entry is $20 a month on both, Codex Plus against Claude Pro. The middle rung is $100 on both, Codex Pro 5x against Claude Max 5x. The top individual rung is $200 on both, Codex Pro 20x against Claude Max 20x. A team seat is $20 per user billed annually and $25 billed monthly on both. Both vendors even describe the step up in the same words, as 5x or 20x more usage than the entry plan. Two companies that compete on almost nothing else have landed on the same four numbers. The practical consequence is that anyone choosing between these tools on price is answering a question that has no answer. Choose on the two things that genuinely differ: what you can find out about the limits before you buy, and which surfaces the agent runs on.
Only one of the two lets you forecast the limits arithmetically
This is the real asymmetry and it is the most useful thing on this page. OpenAI publishes, for every model, the number of local messages you get per rolling five hour window on each plan: GPT-5.6 Sol is 10 to 100 messages on Plus and 50 to 500 on Pro 5x, Terra is 25 to 200 and 125 to 1,000, Luna is 250 to 2,000 and 1,250 to 10,000. It also publishes a credit rate card per million tokens, so you can price a workload before you run it. Anthropic publishes plan prices and a multiplier, and that is the level of detail available. You cannot derive from Anthropic's pricing page how many Claude Code messages a Max 5x plan buys, because the number is not stated. Neither approach is dishonest, and OpenAI's ranges are wide enough that the low end and the high end differ by a factor of ten, which tells you how much task shape matters. But if you need to model spend on a spreadsheet before committing a team, Codex gives you numbers to model with and Claude Code asks you to run a pilot instead.
Anthropic answers the forecasting question a different way, with observed cost
Anthropic does not publish a message table, but it does publish something OpenAI does not, and it is arguably more honest about how this software actually behaves. In its own cost documentation Anthropic states that across enterprise deployments the average is around $13 per developer per active day and $150 to $250 per developer per month, and that costs stay below $30 per active day for 90% of users. That is measured behaviour rather than a theoretical ceiling, and the $150 to $250 figure is worth sitting with, because it sits well above the $100 Max 5x plan and near the $200 Max 20x plan. Read the two vendors together and they are telling you the same thing from opposite directions: sustained agentic coding by a working developer is a $100 to $250 a month activity, whichever logo is on the invoice.
Your coding quota is shared with things that are not coding, on both
Neither product gives the agent a dedicated budget, and this catches teams out. On the OpenAI side, Codex usage is shared with ChatGPT Work, with ChatGPT for Excel on Plus and Pro, with tasks started through Voice, and with image generation, which OpenAI notes drains included limits three to five times faster than an equivalent turn without it. On the Anthropic side, Claude Code shares its allowance with Claude chat and Claude Cowork. So a developer who spends the morning in the chat product arrives at the afternoon's coding session with less headroom, on either tool. If you are budgeting a seat for engineering work, budget for the whole surface the subscription covers, not for the coding agent alone.
Codex has a free floor and Claude Code does not
The cheapest way to try Codex is to not pay: it is included on the $0 ChatGPT plan, with an $8 Go tier between free and Plus. Claude Code starts at Pro, which is $17 a month on annual billing or $20 monthly, and there is no free tier of it at all. For an individual deciding what to learn on, that difference is most of the argument, and it is a real reason Codex adoption spreads faster inside organizations that have not bought anything yet. For a team running a formal evaluation it matters much less, because you were going to buy seats for the pilot either way. Worth knowing which of those two situations you are actually in before treating the free tier as a deciding factor.
Model choice is a philosophical difference, not a feature checkbox
Codex gives you a family to route between and expects you to choose: Sol for hard problems, Terra for everyday production work, Luna when you want speed and volume, with the older GPT-5.5 and 5.4 models still selectable underneath. The rate card makes the trade explicit, since Luna costs five credits per million input tokens against Sol's 125, a twenty five fold difference for the same work if the task is simple enough. Claude Code gives you Anthropic's models and no third party routing, which means fewer decisions and less tuning available. Whether that reads as simplicity or as a constraint depends entirely on whether you enjoy tuning. Teams that want to cut spend by routing cheap tasks to cheap models have a lever on Codex that Claude Code does not offer in the same form.
Codex vs Claude Code for a team: the seat maths
At twenty developers the entry seat costs $400 a month on either side, because Codex Business and Claude Team are both $20 per user billed annually and both $25 billed monthly. There is genuinely nothing to save by switching. What differs is what happens when those twenty developers turn out to use the tool unevenly, which they always do in the first quarter.
Anthropic sells the heavy user as a separate product: a Team premium seat is $100 per seat billed annually, or $125 monthly, and carries five times a standard seat's usage. So you can buy three heavy seats and seventeen standard ones and keep a single predictable invoice. OpenAI handles the same problem with workspace credits, which a Business or Enterprise workspace tops up as needed. The Anthropic model is easier to forecast and the OpenAI model is easier to flex, and which one your finance team prefers is usually a stronger signal than any benchmark.
Budget against Anthropic's own published figure rather than the sticker price either way. It reports around $13 per developer per active day and $150 to $250 per developer per month across enterprise deployments, staying under $30 per active day for 90% of users. That number is measured, not marketed, and it sits above the $100 tier on both products. Our cost per developer breakdown works through how to turn that into a real budget line.
Codex vs Claude Code: which is better for your situation?
"Which is better" has no answer when the prices match. Here is the same question asked nine ways that do.
Swipe to see more
| If this is you | Pick | Why |
|---|---|---|
| You need to model spend on a spreadsheet before you buy seats | Codex | Published per model message ranges and a credit rate card give you numbers to calculate with. |
| You want to try an agent without paying anything first | Codex | It is included on the free ChatGPT plan. Claude Code has no free tier. |
| You want to cut cost by routing simple tasks to a cheaper model | Codex | Luna costs five credits per million input tokens against Sol at 125, and you choose per task. |
| Your developers live in JetBrains IDEs | Claude Code | It ships a JetBrains plugin alongside VS Code, terminal, desktop and web. |
| You want the fewest decisions and the least tuning | Claude Code | One model family, no routing to configure, no rate card to reason about. |
| You already pay for ChatGPT across the company | Codex | The seats you have already bought include it. There is no new line item to approve. |
| You already pay for Claude across the company | Claude Code | Same logic in reverse. Pro, Max and Team seats all include it. |
| You are choosing purely on price | Neither, the question is void | The ladders are identical at $20, $100, $200 and $20 a team seat. There is nothing to choose between. |
| You want the work to arrive as a reviewable pull request on a predictable bill | Worth a third look | Both meter you and both are built to be sat with. See below. |
Running both is a serious answer here in a way it rarely is elsewhere, because the quotas are separate. Codex Plus at $20 plus Claude Pro at $17 annually is $37 a month and gives you two independent five hour windows, which is more total headroom than either $100 tier buys on its own. If you are weighing a wider field, our best AI for coding roundup covers the rest, and Cursor vs Claude Code takes the other half of this bracket.
What both of them leave to you
Codex and Claude Code are both excellent, and they are both tools you sit with. Each can run a long task while you do something else, and each is honest that the meter keeps moving while it does. That is the shared constraint: the more work you delegate, the faster the five hour window empties, and the harder your monthly bill is to predict. Anthropic's own $150 to $250 per developer per month figure is what that looks like at steady state.
Agentcode is built for that gap rather than to replace either one. You describe a task, it plans, edits, runs your test suite, and opens a reviewable pull request on the GitHub or GitLab repository you already have. It never merges on its own. The bill is $29 a month, flat, with no credit rate card and no five hour window, so neither table on this page has a row for us: there is nothing to meter.
Where they genuinely beat us: both ship far broader model access than we do, both have interactive editing loops we do not attempt, and Codex has a free tier we cannot match. If you want a frontier model at your fingertips while you type, buy one of them. If you want changes arriving as reviewable pull requests on GitLab as well as GitHub at a price you can forecast a year out, that is the case for us. Compare directly on Claude Code alternatives.
Codex vs Claude Code: your questions answered
Is Codex better than Claude Code?
Neither is better outright, and price cannot separate them because both ladders cost $20, $100 and $200. Codex is better if you want to forecast usage from published numbers, route tasks between cheap and expensive models, or start free. Claude Code is better if you want one model family with nothing to tune, a JetBrains plugin, or you already pay for Claude across the company.
Which is cheaper, Codex or Claude Code?
They cost the same at every comparable rung. Entry is $20 a month on both, Codex Plus against Claude Pro. The 5x tier is $100 on both and the 20x tier is $200 on both. A team seat is $20 per user billed annually and $25 monthly on both. The only price difference is below the entry rung, where Codex is included free and offers an $8 Go tier, and Claude Code has no equivalent.
What are the usage limits for Codex vs Claude Code?
Both run a rolling five hour window with weekly limits on top. OpenAI publishes the actual numbers per model: on Plus you get 10 to 100 GPT-5.6 Sol messages per five hour window, 25 to 200 on Terra and 250 to 2,000 on Luna, multiplied by five on Pro 5x. Anthropic publishes a multiplier rather than message counts, so Claude Code limits have to be established by running a pilot.
Codex vs Claude Code: which is better for a team?
The seat price is identical at $20 per user billed annually, so decide on governance and forecasting. Codex lets finance model spend from a published rate card and lets you route cheap work to cheap models. Anthropic sells a distinct premium seat at $100 annually for heavy users and publishes observed cost data, around $13 per developer per active day, which is useful for a budget case.
How much does Codex cost per month?
Codex is included on the free ChatGPT plan at $0, on Go at $8 a month, and on Plus at $20. Pro starts at $100 a month for 5x the Plus rate limits, with a $200 tier for 20x. Business is $20 per user per month billed annually for two or more users, or $25 billed monthly, and Enterprise is custom. You can also sign in with an API key and pay standard API rates instead.
How much does Claude Code cost per month?
Claude Code is included with a paid Claude plan. Pro is $17 a month billed annually or $20 monthly. Max is $100 a month for 5x Pro usage and $200 for 20x. A Team standard seat is $20 billed annually or $25 monthly, a Team premium seat is $100 annually or $125 monthly for 5x a standard seat, and Enterprise is $20 a seat plus usage at API rates.
Can you use Codex and Claude Code together?
Yes, and at the entry rung it is cheaper than most people assume. Codex Plus at $20 plus Claude Pro at $17 on annual billing comes to $37 a month, well under a single $100 tier of either. They are separate CLIs and separate subscriptions with separate quotas, so running both also doubles your headroom before either five hour window runs out.
Does Codex or Claude Code have a free version?
Codex does. It is included on the $0 ChatGPT Free plan, with limits well below the paid tiers, and an $8 Go plan sits between free and Plus. Claude Code has no free tier: it requires at least a Claude Pro subscription at $17 a month billed annually. If cost of trial is what you are optimizing for, that is the one structural difference between the two lineups.
What models does Codex use?
The GPT-5.6 family. Sol handles complex reasoning, ambiguous problems and advanced coding, Terra is the everyday production workhorse, and Luna is optimized for fast high volume work. GPT-5.5, GPT-5.4 and GPT-5.4 mini remain selectable at lower credit rates. Pro users also get GPT-5.3-Codex-Spark, a fast research preview model governed by its own separate usage limit.
Why did my Codex or Claude Code limit run out so fast?
Usually because something other than coding spent it. Codex shares its allowance with ChatGPT Work, ChatGPT for Excel, Voice initiated tasks and image generation, which OpenAI says drains limits three to five times faster. Claude Code shares its allowance with Claude chat and Cowork. On Codex, fast mode also consumes credits at a higher rate, and long sessions cost more because the agent holds more context.
Last updated: August 2026
Still comparing?
Full pricing breakdowns for each tool: Claude Code pricing, Cursor pricing, GitHub Copilot pricing, Kiro pricing and Devin pricing.
Or read why Codex and Claude Code limits run out, Cursor vs Claude Code, Cursor vs GitHub Copilot, Claude Code vs GitHub Copilot pricing, or step back to AI coding tool pricing across every tool and what an AI coding agent actually does.
A flat bill, on your repo
Connect a GitHub or GitLab repository you already have, describe a task, and read the pull request it opens. No five hour window to watch and no credits to forecast.