Introduction
OpenAI Codex pricing now stretches from a $0 Free plan to a $200-per-month Pro 20x plan. Between them sit Go at $8, Plus at $20, and Pro 5x at $100. The list prices are easy to compare. The harder question is what you are actually buying, because 5x and 20x describe Codex usage relative to Plus—not five or twenty times better code, and not unlimited access.
This guide explains the current plans, shared usage windows, extra credits, Fast mode, API billing, and the cost of a real coding task. It also separates Codex from Yukie. Codex is designed for working with code through surfaces including the web, terminal CLI, and IDE. Yukie is a dedicated web-based agent workspace within GlobalGPT. They overlap as agents, but Yukie is not a Codex CLI add-on and should not be presented as one.
OpenAI Codex Pricing at a Glance
The prices and plan descriptions above were verified against OpenAI’s official Codex pricing documentation on July 27, 2026. Because product pricing and limits can change, recheck that page before subscribing. If you are comparing more than price, our guide to the beste AI-Modelle für die Codierung explains why model quality, tool access, and workflow fit matter as much as the monthly fee.
What Do You Get With Each Codex Plan?
The paid tiers are not simply the same plan at higher prices. Free is for evaluation, Go covers lightweight use, and Plus is the first tier with an explicitly documented cross-device Codex experience. Here is where each one fits.
Codex Free — Best for Trying It
Codex Free costs $0 per month. OpenAI positions it for exploring quick coding tasks, which makes it the sensible starting point when you are still learning what an AI coding agent can do. Use it to test a bounded request: explain a small function, review a short patch, or propose a unit test.
Do not assume Free offers the same capacity or complete set of surfaces as Plus. OpenAI’s current price card confirms the price and intended use, but does not make the same explicit web, CLI, IDE, and iOS promise shown for Plus. Treat Free as an evaluation tier, not as guaranteed daily development capacity. A broader ChatGPT Go vs. kostenlose Version – Vergleich can help if your decision also includes general ChatGPT use.
Codex Go — The $8 Lightweight Option
Codex Go costs $8 pro Monat and is positioned for lightweight coding tasks. It fills the gap between occasional free use and the $20 Plus plan. That makes it attractive to learners, side-project builders, and users who want more room than Free without paying for a full Plus subscription.
Go should be judged by whether it covers your normal tasks before a reset or limit interrupts you. Its low monthly price does not guarantee enough capacity for large repositories or long autonomous sessions. For a closer look at the wider subscription, read our ChatGPT Go Bewertung.
Codex Plus — The $20 Starting Point
Codex Plus costs $20 pro Monat. OpenAI describes it as suitable for a few focused coding sessions each week and explicitly includes Codex on the web, in the CLI, in the IDE extension, and on iOS. For many individual developers, this is the first tier that combines predictable billing with a complete Codex workflow.
Plus is not a bucket of fixed tokens, and the number of tasks you can finish is not guaranteed. Model choice, context size, reasoning, tool use, retrieval, and caching all affect consumption. If your work repeatedly stops at the limit, compare the interruption cost with a higher plan rather than upgrading because “Pro” sounds more capable. Our guide to ChatGPT Plus Grenzen explains the same budgeting principle in more detail.
Codex Pro Pricing: 5x vs 20x
The Pro decision is mainly about how much uninterrupted capacity you need. Both tiers include the Plus experience, but their usage multiples and monthly prices are very different.
Pro 5x Costs $100 per Month
Codex Pro 5x costs $100 per month and provides five times the Codex usage available on Plus. The “5x” label refers to usage or rate limits. It does not mean the same model becomes five times more accurate, produces code five times faster, or solves problems five times as difficult.
Pro 5x makes economic sense when Plus limits interrupt paid work often enough to cost more than the $80 monthly upgrade. It can also be easier to budget than open-ended API billing. But if your coding work is occasional, paying five times the Plus price for unused capacity is poor value.
Pro 20x Costs $200 per Month
Codex Pro 20x costs $200 pro Monat and provides twenty times the Codex usage available on Plus. It is aimed at heavy individual use: repeated repository analysis, sustained implementation work, parallel tasks, and workflows where a usage reset would interrupt a normal workday.
Twenty times Plus usage still does not mean unlimited Codex. Local messages and cloud chats share a five-hour window, additional weekly limits may apply, and resource-intensive modes can consume that allowance faster.
Is Pro 20x Worth Twice the Price?
Compared with Pro 5x, Pro 20x doubles the monthly bill from $100 to $200 but increases the stated usage multiple from 5x to 20x. On paper, that is four times the Plus-relative capacity for twice the Pro 5x price. In practice, it is worth it only if you regularly use the added allowance.
If your real question is which model to use inside a coding workflow, see our comparison of the best ChatGPT models for coding. A higher subscription allowance and a better model choice solve different problems.
The Real Cost of Codex
The subscription fee is only the starting point. To understand what Codex really costs, you also need to account for usage windows, task complexity, faster-consuming features, and the price of correcting an unsuccessful run.
The Monthly Price Is Not an Unlimited Pass
A Codex subscription prepays for access to a limited allowance. OpenAI states that local messages and cloud chats share a five-hour usage window and that additional weekly limits may apply. A task completed inside Plus or Pro may produce no separate line-item charge, but it still consumes part of the plan you already paid for.
This distinction matters when someone says a subscription task “cost $0.” A more accurate statement is: no incremental charge was observed because the task was included in the subscription. The economic cost can be allocated across the successful work completed that month.
These figures are personal cost allocations, not OpenAI’s per-task prices. They also assume every counted task produced an accepted result. A failed run that needs extensive human correction increases the true cost even if it used fewer tokens.
Different Tasks Consume Different Amounts
Codex does not charge every task equally. A request to rename one variable may touch a small amount of context. A request to trace a production bug can require repository search, file reads, repeated reasoning, tool calls, test output, and multiple revisions. Prompt length alone is therefore a poor cost estimate.
- Modellwahl: different models have different subscription consumption and API rates.
- Context size: more files and longer conversations require more input processing.
- Begründung: harder tasks may consume more compute or credits.
- Tool use and retrieval: searches, commands, test output, and external retrieval add context.
- Caching: reusable context can be priced or counted differently from fresh input.
- Wiederholungsversuche: a cheap failed attempt can cost more overall than a slightly heavier first-run success.
Some Features Use Limits Faster
OpenAI says image generation uses the same general limits and, on average, consumes included limits three to five times faster than a similar turn without image generation, depending on image quality and size. Free does not include that capability. With an API key, image use follows API pricing instead.
Fast mode is another tradeoff. OpenAI’s current speed documentation says it makes supported models about 1.5x faster, but GPT-5.6 and GPT-5.5 consume credits at 2.5x the Standard rate, while GPT-5.4 consumes credits at 2x. Faster output can be valuable during urgent work, but it is not a free performance upgrade.
What Happens When You Reach the Limit?
When included usage runs out, the lowest-cost response is to wait for the limit to reset. Plus and Pro users can also extend usage with available ChatGPT credits. Those credits are separate from the included allowance; do not assume they are automatically included for free.
In an active Codex CLI session, run /status to check remaining limits. The Codex usage dashboard shows the same information outside the session. Check before starting a large task, especially if you plan to use Fast mode, image generation, or a long context.
You can also use an API key in the CLI, SDK, or IDE extension. That route has no fixed Codex subscription price: it bills the selected model’s token usage at current API rates and does not include Codex cloud features such as GitHub code review or Slack integrations. Subscription usage and API billing should never be treated as one combined balance.
How Much Does a Real Codex Task Cost?
Testing status: We designed a controlled Codex-versus-Yukie test, but have not yet run the paid sessions. The method below is ready; the token, credit, billing, and timing results are not measured. Missing values are not estimated or presented as observed data.
A useful task-cost comparison must hold the repository, prompt, environment, time limit, and acceptance tests constant. The sections below show the task and the measurements required before either agent can be called the better value.
The Test Task
The benchmark uses a small frozen Python 3.12 repository with one date-parsing bug. Each agent receives the same clean repository copy and the same prompt:
Find and fix a date-parsing bug so the public function accepts ISO 8601 timestamps ending in Z or containing an explicit numeric UTC offset, rejects timestamps without timezone information, normalizes valid results to UTC, adds regression tests, runs the complete test suite, and explains the changes. Do not install dependencies, access the network, or modify unrelated files.
The public acceptance checks cover Z, positive offsets, missing timezones, invalid input, and timezone-aware UTC output. Hidden checks cover a negative offset and fractional seconds. Each run gets the same 15-minute limit, no mid-task human hints, and independent test verification. If a first run fails, that result remains recorded; any correction is counted separately.
How Many Tokens Did Codex Use?
No measured token total is available yet. The benchmark will record the exact model and version, input tokens, cached input tokens, cache writes, output tokens, total tokens, credits consumed, and subscription allowance change. If Codex does not expose one of those fields, the result will say “Not exposed by the provider” instead of estimating it from message length.
For API use, the comparison will apply the rates for the exact model used. OpenAI’s current Standard rates per one million tokens are shown below for reference; they are not observed costs for this unrun task.
Rates verified on July 27, 2026. Requests above documented long-context thresholds may use different rates; always check Preise für die OpenAI-API before calculating.
What Did the Codex Task Really Cost?
The answer depends on the payment route. Under a ChatGPT subscription, the benchmark will report the observed change in included usage and whether any extra credits were deducted. If no separate charge appears, the accurate result is “no incremental charge observed; included in the subscription,” not “the task was free.”
An API-equivalent cost is only a reference estimate based on token pricing. It is not an extra per-task charge for work already covered by a ChatGPT subscription. The benchmark will keep three figures separate: actual billed cost, subscription quota consumed, and API-equivalent cost.
How Much Did the Same Task Cost on Yukie?
The same-task Yukie cost has not yet been measured. The test will record the exact model or mode shown, credits consumed, balance change, elapsed time, test results, manual correction turns, and whether an accepted result was produced.
GlobalGPT’s public pricing page currently displays annual-plan monthly equivalents of $5.80 for Basic, $10.80 for Pro, and $25 for Unlimited, all billed annually. Those platform prices are far below Codex Pro 5x and Pro 20x. However, the public page does not state which plan includes Yukie or disclose Yukie-specific task, token, or credit limits. Until that mapping is confirmed, these numbers must be described as GlobalGPT platform prices—not as verified Yukie-only pricing.
Which Agent Delivered Better Value?
No test winner can be named before both agents produce independently verified results. The comparison will prioritize cost per accepted result, not the cheapest first attempt. It will report actual billed cost, elapsed time, public and hidden test pass rates, first-run success, human correction turns, and unrelated files changed.
Yukie should only be called cheaper for this coding task if it meets the same acceptance criteria at a lower observed cost and without significantly more correction. A lower subscription price is promising, but a failed result is not better value.
A Cheaper Codex Alternative: When Yukie Makes More Sense
Codex is built for coding and developer tools. Yukie is the simpler, lower-cost choice when a browser agent can complete the job.
Yukie Is an Independent Agent
Yukie is a dedicated web-based GlobalGPT agent for images, videos, documents, research, tables, websites, and slides. It works independently; it is not a Codex CLI plugin.
Lower Cost Without the Full Codex Bundle
GlobalGPT’s annual-plan prices start far below Codex Pro, making Yukie a practical option for web-first tasks. Yukie’s exact plan eligibility still needs confirmation.
Unser all-in-one AI platform comparison covers more lower-cost alternatives.
Less Setup for Non-Developers
Yukie starts in the browser, so non-developers can skip repository, command-line, and IDE setup. That can save both subscription cost and setup time; see our AI assistant comparison for the same outcome-first approach.
When Yukie May Not Replace Codex
Choose Codex when you need direct repository editing, terminal commands, tests, IDE integration, GitHub review, or other Codex cloud features. Yukie is not a verified replacement for those developer workflows.
Codex vs Yukie: What Is the Difference?
The comparison is not “which agent is universally better?” It is “which agent completes this particular job at an acceptable quality and total cost?” Codex earns its higher price when direct repository access and developer tools save substantial time. Yukie has the lower-cost advantage when a browser agent is enough.
Final Verdict — Do You Really Need Codex Pro?
Codex Plus is the logical starting point for someone who needs the complete web, CLI, IDE, and iOS workflow. Pro 5x makes sense for frequent users who outgrow Plus. Pro 20x is a specialist tier for heavy use where waiting has a measurable business cost. API billing is useful when you need programmatic or variable usage, but it must be tracked separately.
If you do not need local repository control, a terminal, or an IDE agent, do not pay for them by default. Yukie is an independent browser agent, not a Codex extension, and its surrounding GlobalGPT plans are priced far below Codex Pro. For research, media, documents, websites, slides, and other web-first outcomes, it may be the more economical choice once the applicable Yukie plan and limits are confirmed.
The final price decision should follow one rule: compare the cost of an accepted result, including correction time, rather than the cheapest subscription label. The controlled Codex-versus-Yukie task above must be completed before making a measured winner claim.
Häufig gestellte Fragen
These answers cover the pricing, limits, billing routes, and Yukie distinction that matter most before choosing a plan.
Is OpenAI Codex free?
Yes. OpenAI currently lists Codex Free at $0 per month for exploring quick coding tasks. Its capacity is lower than paid plans, and the current price card does not promise the same complete surface coverage as Plus.
How much does Codex Plus cost?
Codex Plus costs $20 per month. OpenAI says it includes Codex on the web, in the CLI, in the IDE extension, and on iOS.
How much does Codex Go cost?
Codex Go costs $8 per month and is positioned for lightweight coding tasks. It sits between the $0 Free plan and $20 Plus plan.
What is Codex Pro 5x?
Codex Pro 5x is a $100-per-month plan that provides five times the Codex usage available on Plus. The 5x label refers to usage capacity, not five times better model quality.
What is Codex Pro 20x?
Codex Pro 20x costs $200 per month and provides twenty times the Codex usage available on Plus. It is designed for heavy individual use but is not unlimited.
Is Pro 20x twice as good as Pro 5x?
No. Pro 20x doubles the monthly price and offers a larger usage allowance, but it does not make the model twice as capable. Quality still depends on the model, task, context, and instructions.
Is Codex Pro unlimited?
No. Local messages and cloud chats share a five-hour usage window, and additional weekly limits may apply. Resource-intensive tasks can consume the allowance faster.
Does a Codex subscription include API usage?
No. API-key access is billed separately by token at the selected model’s API rates. It works in the CLI, SDK, and IDE extension but does not include Codex cloud features.
How can I check my Codex usage?
Ausführen /status in an active Codex CLI session to see remaining limits, or open the Codex usage dashboard outside the session.
What happens when I reach a Codex limit?
You can wait for the allowance to reset. Plus and Pro users may also continue with available ChatGPT credits, which are separate from included usage.
Does Codex Fast mode cost more?
It consumes credits faster. OpenAI says Fast mode is about 1.5x faster on supported models; GPT-5.6 and GPT-5.5 use credits at 2.5x the Standard rate, while GPT-5.4 uses credits at 2x.
Is Yukie cheaper than Codex?
GlobalGPT’s current annual-plan monthly equivalents are lower than paid Codex prices, but the public pricing page does not specify which plan includes Yukie or its exact limits. Yukie can be a lower-cost option for suitable browser-agent tasks, but a direct per-task savings claim requires verified access and benchmark data.
Does Yukie run inside the Codex CLI?
No. Yukie is an independent web-based agent workspace within GlobalGPT. It is an alternative for some outcomes, not a Codex CLI plugin or an agent that must be connected to Codex.
