Codex Pricing in 2026: What Free, Plus, Pro, Business, and API Really Cost
OpenAI Codex pricing in 2026: Free, Go, Plus, Pro, Business, Enterprise, API rates, five-hour limits, and which plan builders should buy.
Codex costs $0, $8, $20, $100, $200, or API token rates depending on how hard you push it. The $20 Plus plan is the right default for most solo builders; Pro only earns the upgrade when you regularly run out of the five-hour window.
OpenAI Codex is not a separate subscription. It is bundled into ChatGPT plans, then gated by usage limits, credits, or raw API token billing depending on how you access it.
The trap is reading the monthly price as the whole cost. Codex pricing has two layers: the plan you buy and the amount of agent work your tasks consume. A file-level bug fix and a repo-wide refactor do not spend the same allowance, even if both look like a single message.
At a glance
The short answer: start on Plus, not Pro
Plus is the first serious Codex plan because it includes the surfaces builders actually use: Codex on the web, the CLI, the IDE extension, and iOS. It also includes cloud-based integrations such as automatic code review and Slack integration, plus access to GPT-5.5, GPT-5.4, and GPT-5.4 mini.
That is enough for a founder fixing a bug, an operator changing an internal tool, or a solo developer giving Codex scoped tasks between normal coding sessions. The $20 plan gets expensive only when it creates stop-start work. If Codex pauses in the middle of a multi-file change and you spend an hour waiting for the window to reset, the real cost is not $20. It is the lost session.
Pro starts at $100/month and gives 5x more usage than Plus. The $200 Pro tier gives 20x more usage than Plus. That sounds like a power-user ladder, but it is better treated as an interruption tax. Do not buy Pro because you feel like a serious developer. Buy it when Plus starts cutting off serious work.
- Plus is cheap enough to be the default for most individual builders.
- Pro 5x has a clear job: protect long coding sessions from usage interruptions.
- Pro 20x is the heavy-work tier for parallel sessions and all-day agent use.
- API key usage is available when you need automation outside ChatGPT plans.
- The plan price does not tell you how much work one task will consume.
- Free and Go are trials for coding, not dependable daily-agent plans.
- Business seats can still behave like Plus seats unless flexible pricing is enabled.
- API key usage loses cloud features such as GitHub code review and Slack integration.
If you are comparing Codex with Claude Code, use the dedicated Codex vs Claude Code breakdown. This page is the Codex pricing decision by itself: which plan to buy, what the limits mean, and when token billing matters.
Codex pricing table: what each plan buys
OpenAI's Codex pricing page splits Codex into three buying modes: personal ChatGPT plans, workspace plans, and API-key usage. The personal ladder is simple on the surface: Free at $0, Go at $8, Plus at $20, Pro 5x at $100, and Pro 20x at $200.
Free and Go are useful for proving whether Codex fits your workflow. They are not the plan to choose if you expect Codex to work through real project changes each week. Go at $8 is cheap, but "lightweight coding tasks" is the phrase that matters. Treat it as a low-friction test, not as a production coding budget.
Plus at $20/month is the first paid plan that maps to normal solo work. It gives you the local and web surfaces, plus the integration layer that makes Codex more than a terminal command. If your work is scoped, such as "add a form validation test", "trace this regression", or "update this component to match the new API", Plus is the right first buy.
Pro changes the shape of the day. The $100 tier gives 5x more usage than Plus. The $200 tier gives 20x more usage than Plus. Both include the same core Pro capabilities, and the higher tier is about usage allowance, not a different kind of Codex.
Business is not simply "Plus for teams." It brings workspace controls, SAML SSO, MFA, billing controls, larger virtual machines for cloud tasks, and a no-training-by-default posture for business data. The important pricing catch is that Business is $20/user/month when billed annually, but $25/user/month when billed monthly, and the listed local-message ranges match Plus unless flexible pricing is enabled.
API-key usage is a different product shape. It gives you Codex in the CLI, SDK, or IDE extension and bills by tokens. It does not include cloud-based features such as GitHub code review or Slack integration, and model availability follows the models available to your API key.
The real limit is the five-hour window
The five-hour window decides how Codex feels in practice. Plus can feel generous for scoped changes, then feel too small when a large refactor burns through the allowance.
OpenAI lists usage ranges because a "message" is not a fixed unit. A small script may consume a fraction of the allowance. A larger codebase, a long-running task, or a session that needs more context can consume much more. Codex usage also counts toward agentic usage alongside features such as ChatGPT for Excel and Workspace Agents.
The wide ranges are the point. Codex is not billing your intent. It is consuming compute. A clean task in a small repo may fit near the high end of the range. A vague instruction in a large repo can land near the low end because Codex reads more files, carries more context, and loops longer.
Think of the window like a workbench reservation. Plus gives you a few focused sessions. Pro 5x gives you enough bench time to keep working through a heavier block. Pro 20x is for the person who keeps multiple benches active or runs Codex throughout the day.
Cloud tasks and local messages share the same five-hour window where applicable, and additional weekly limits may apply. If you hit a limit during an active turn, Codex can continue that turn subject to fair-use limits. After that, Plus and Pro users may buy credits, wait for reset, or switch down to a smaller model.
When Pro is worth the upgrade
Pro is worth it when Codex interrupts real work, not when the spec sheet looks better. The $100 tier gives 5x Plus usage. The $200 tier gives 20x Plus usage. Both tiers include Pro capabilities, and both are intended for people who rely on Codex for heavier sessions.
Use this decision rule:
For a solo technical founder, Plus is the default. If you are building a feature after dinner, fixing bugs between calls, and asking Codex for bounded changes, the $20 plan is hard to beat.
For a senior engineer using Codex as a real pair during implementation blocks, Pro 5x is the first serious upgrade. The upgrade pays for itself when it protects deep sessions that would otherwise stall.
For an agency owner or internal-tools builder running Codex across several repos, Pro 20x can be rational. The mistake is buying it before the usage pattern exists. If Plus is not interrupting you, Pro 20x is just unused capacity.
If you are deciding between OpenAI and Anthropic plans, pair this with the Claude Code pricing breakdown. The right comparison is not $20 versus $20. It is how often each agent stops your work, what features you lose by changing tools, and which model is better for your codebase.
When API pricing matters
Use OpenAI's API pricing page when Codex is part of automation instead of a human ChatGPT session. A developer using the CLI with a ChatGPT login is usually choosing a plan. A team wiring Codex into CI, a custom SDK workflow, or shared automation should look at token billing.
The API path is cleaner for machines and worse for casual humans. It bills what you use, which is exactly what you want for a CI pipeline that runs only on pull requests. It is less pleasant when you are a solo builder who just wants a predictable monthly price and the ChatGPT cloud features.
API-key Codex also has a product tradeoff. It works in the CLI, SDK, or IDE extension, but it does not include cloud-based features such as GitHub code review or Slack integration. If those features are why you want Codex, stay inside the ChatGPT plan path.
GPT-5.3-Codex-Spark adds another wrinkle. It is a Pro research preview model for day-to-day coding tasks and is not available in the API at launch. If Spark is the reason you want Codex, the API is not the path.
How to make your Codex limit last longer
Most Codex limit problems start as scope problems. Before upgrading, reduce the amount of context Codex has to carry.
Give Codex a smaller job
Ask for a concrete change at a time: "add tests for this function", "fix this failing route", or "explain why this migration fails". Broad instructions make Codex inspect more files and spend more of the window.
Switch routine work to GPT-5.4 mini
OpenAI lists GPT-5.4 mini at 60-350 local messages per five-hour window on Plus, compared with 15-80 for GPT-5.5. Use the smaller model for routine edits, boilerplate, and low-risk refactors.
Trim AGENTS.md and prompt context
Long project instructions help only when they are relevant. If a repo-level AGENTS.md injects a wall of process into every task, split guidance into nested files so small tasks do not carry the whole company handbook.
Disable unused MCP servers
Every connected tool can add context. Keep the MCP servers you need for the task and turn off the ones that are just available by habit.
Avoid Fast mode unless speed is worth the burn
Speed configurations increase credit consumption, and Fast mode uses credits at a higher rate for supported models. Use it when turnaround matters more than allowance.
Do image work outside Codex when coding allowance matters
Image generations use included limits 3-5x faster on average than similar turns without image generation. If the goal is code, keep visual generation out of the Codex session.
These steps do not make Plus behave like Pro. They make sure you upgrade for real demand rather than messy prompts.
Who should pick which Codex plan
Pick the cheapest plan that does not interrupt the job you actually do. Codex is valuable when it shortens the path from intention to working code. It is annoying when the plan becomes a scheduling constraint.
The blunt rule: Plus is the default, Pro 5x is the working-builder upgrade, Pro 20x is the heavy-usage tier, and API is for systems. Anything else is overbuying.
The plans to avoid
Free is the plan to avoid for real work. It is useful for a quick look, but it is not a dependable build setup.
Go is easy to overestimate. At $8 it looks like a bargain, but it is positioned for lightweight coding tasks. If Codex is going to touch real project code every week, Plus is the cleaner baseline.
Business is the wrong buy for a single independent builder. Its value is workspace control: billing, admin, SSO, MFA, usage analytics, no training on business data by default, and larger virtual machines for cloud tasks. If you do not need those, buy a personal plan.
API-key usage is the wrong buy when you want the ChatGPT Codex product experience. It is powerful, but it is not a replacement for cloud code review, Slack, and the bundled ChatGPT plan surfaces.
If none of the Codex plans fit, the next question is not "which Codex tier is hidden?" It is whether another coding agent matches your workflow better. The Claude Code alternatives guide is the better next read when the blocker is fit rather than price.
How much does Codex cost?
Codex is bundled into ChatGPT plans: Free is $0/month, Go is $8/month, Plus is $20/month, Pro is $100/month for 5x usage or $200/month for 20x usage, Business starts at $20/user/month annually or $25 monthly, Enterprise and Edu are custom, and API-key usage bills by tokens.
Is Codex free or paid?
Both. Free includes limited Codex access, but serious daily coding usually starts at Plus for $20/month because it includes expanded Codex usage across the web, CLI, IDE extension, and iOS.
Is Codex free with ChatGPT Plus?
Yes. ChatGPT Plus includes Codex, and the Codex pricing page lists Plus at $20/month with expanded access across web, CLI, IDE extension, and iOS.
Is Codex free with ChatGPT Go?
Go includes Codex access for lightweight coding tasks at $8/month. It is cheaper than Plus, but Plus is the better baseline if Codex will touch real project work every week.
Is Codex cheaper than Claude Code?
For light or moderate solo work, Codex Plus at $20/month is inexpensive. For heavy agent use, the comparison depends on interruption frequency, model fit, and whether you need Pro 5x, Pro 20x, or API billing.
Is Codex really worth it?
Codex is worth it when the plan saves more time than it costs. If one debugging loop, test-writing pass, or scoped refactor saves an hour in a month, Plus has paid for itself. If it stops you mid-task most workdays, Pro 5x is the more honest price.
How much does Codex API usage cost?
The specialized gpt-5.3-codex API rate is $1.75 per 1M input tokens, $0.175 per 1M cached input tokens, and $14.00 per 1M output tokens on standard processing. Priority processing is $3.50 input, $0.35 cached input, and $28.00 output per 1M tokens.
What happens when I hit a Codex usage limit?
If Codex reaches a usage limit during an active turn, it can continue that turn subject to fair-use limits. After that, Plus and Pro users can buy credits, switch to a smaller model, upgrade, or wait for the limit to reset.
Jul 8, 2026
